Menu
▾
▴
kaldi-users — Kaldi Users
You can subscribe to this list here.
| 2011 |
Jan
|
Feb
|
Mar
|
Apr
|
May
|
Jun
|
Jul
(2) |
Aug
(2) |
Sep
(1) |
Oct
(1) |
Nov
|
Dec
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2012 |
Jan
|
Feb
|
Mar
(8) |
Apr
(4) |
May
(2) |
Jun
(1) |
Jul
|
Aug
|
Sep
|
Oct
|
Nov
|
Dec
|
| 2013 |
Jan
|
Feb
(2) |
Mar
(2) |
Apr
(7) |
May
(31) |
Jun
(40) |
Jul
(65) |
Aug
(37) |
Sep
(12) |
Oct
(57) |
Nov
(15) |
Dec
(35) |
| 2014 |
Jan
(3) |
Feb
(30) |
Mar
(57) |
Apr
(26) |
May
(49) |
Jun
(26) |
Jul
(63) |
Aug
(33) |
Sep
(20) |
Oct
(153) |
Nov
(62) |
Dec
(20) |
| 2015 |
Jan
(6) |
Feb
(21) |
Mar
(42) |
Apr
(33) |
May
(76) |
Jun
(102) |
Jul
(39) |
Aug
|
Sep
|
Oct
|
Nov
|
Dec
|
|
From: Dimitris V. <dva...@gm...> - 2015-05-24 08:09:49
|
We have access to several corpora and we are trying to put together something appropriate. In the next couple of days, we will also volunteer a server to set it all up and run the tests. Dimitris > On 24 Μαΐ 2015, at 02:06, Daniel Povey <dp...@gm...> wrote: > > One possibility is to use a completely open-source setup, e.g. > Voxforge, and forget about the "has a clear advantage" requirement. > E.g. target anything that looks like a year, and make a grammar for > years. > Dan > > > On Fri, May 22, 2015 at 6:32 AM, Nagendra Goel > <nag...@go...> wrote: >> Since I cannot volunteer my enviornment, do you recommend another >> enviornment where this can be prototyped and where you can check in some >> class lm recipe that has advantage. >> >> Nagendra >> >> Nagendra Kumar Goel >> >>> On May 21, 2015 11:01 PM, "Dimitris Vassos" <dva...@gm...> wrote: >>> >>> +1 for the class-based LMs. I have also been interested in this >>> functionality for some time now, so will be more than happy to try out the >>> current implementation, if possible. >>> >>> Thanks >>> Dimitris >>> >>>> On 22 Μαΐ 2015, at 01:34, kal...@li... >>>> wrote: >>>> >>>> Send Kaldi-users mailing list submissions to >>>> kal...@li... >>>> >>>> To subscribe or unsubscribe via the World Wide Web, visit >>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >>>> or, via email, send a message with subject or body 'help' to >>>> kal...@li... >>>> >>>> You can reach the person managing the list at >>>> kal...@li... >>>> >>>> When replying, please edit your Subject line so it is more specific >>>> than "Re: Contents of Kaldi-users digest..." >>>> >>>> >>>> Today's Topics: >>>> >>>> 1. Re: LM grafting (Daniel Povey) >>>> 2. Re: LM grafting (Kirill Katsnelson) >>>> 3. Re: LM grafting (Hainan Xu) >>>> 4. Re: LM grafting (Sean True) >>>> >>>> >>>> ---------------------------------------------------------------------- >>>> >>>> Message: 1 >>>> Date: Thu, 21 May 2015 15:04:04 -0400 >>>> From: Daniel Povey <dp...@gm...> >>>> Subject: Re: [Kaldi-users] LM grafting >>>> To: Sean True <se...@se...> >>>> Cc: Hainan Xu <hai...@gm...>, >>>> "kal...@li..." >>>> <kal...@li...>, Kirill Katsnelson >>>> <kir...@sm...> >>>> Message-ID: >>>> <CAEWAuySHaXwdNJZAoL6CanzHth=k4Y...@ma...> >>>> Content-Type: text/plain; charset=UTF-8 >>>> >>>> The general approach is to create an FST for the little language >>>> model, and then to use fstreplace to replace instances of a particular >>>> symbol in the top-level language model, with that FST. >>>> The tricky part is ensuring that the result is determinizable after >>>> composing with the lexicon. In general our solution is to add special >>>> disambiguation symbols at the beginning and end of each of the >>>> sub-FSTs, and of course making sure that the sub-FSTs are themselves >>>> determinizable. >>>> Dan >>>> >>>> >>>>> On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> >>>>> wrote: >>>>> That's a subject of some general interest. Is there a discussion of the >>>>> general approach that was taken somewhere? >>>>> >>>>> -- Sean >>>>> >>>>> Sean True >>>>> Semantic Machines >>>>> >>>>>> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> >>>>>> wrote: >>>>>> >>>>>> Nagendra Goel has worked on some example scripts for this type of >>>>>> thing, and with Hainan we were working on trying to get it cleaned up >>>>>> and checked in, but he's going for an internship so it will have to >>>>>> wait. But Nagendra might be willing to share it with you. >>>>>> Dan >>>>>> >>>>>> >>>>>> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson >>>>>> <kir...@sm...> wrote: >>>>>>> Suppose I have a language model where one token (a "word") is a >>>>>>> pointer >>>>>>> to a whole another LM. This is a practical case when you expect an >>>>>>> abrupt >>>>>>> change in model, a clear example being "my phone number is..." and >>>>>>> then >>>>>>> you'd expect them rattling a string of digits. Is there any support >>>>>>> in kaldi >>>>>>> for this? >>>>>>> >>>>>>> Thanks, >>>>>>> >>>>>>> -kkm >>>>>>> >>>>>>> >>>>>>> ------------------------------------------------------------------------------ >>>>>>> One dashboard for servers and applications across >>>>>>> Physical-Virtual-Cloud >>>>>>> Widest out-of-the-box monitoring support with 50+ applications >>>>>>> Performance metrics, stats and reports that give you Actionable >>>>>>> Insights >>>>>>> Deep dive visibility with transaction tracing using APM Insight. >>>>>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>>>>>> _______________________________________________ >>>>>>> Kaldi-users mailing list >>>>>>> Kal...@li... >>>>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >>>>>> >>>>>> >>>>>> >>>>>> ------------------------------------------------------------------------------ >>>>>> One dashboard for servers and applications across >>>>>> Physical-Virtual-Cloud >>>>>> Widest out-of-the-box monitoring support with 50+ applications >>>>>> Performance metrics, stats and reports that give you Actionable >>>>>> Insights >>>>>> Deep dive visibility with transaction tracing using APM Insight. >>>>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>>>>> _______________________________________________ >>>>>> Kaldi-users mailing list >>>>>> Kal...@li... >>>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >>>> >>>> >>>> >>>> ------------------------------ >>>> >>>> Message: 2 >>>> Date: Thu, 21 May 2015 19:24:38 +0000 >>>> From: Kirill Katsnelson <kir...@sm...> >>>> Subject: Re: [Kaldi-users] LM grafting >>>> To: "dp...@gm..." <dp...@gm...>, Sean True >>>> <se...@se...> >>>> Cc: Hainan Xu <hai...@gm...>, >>>> "kal...@li..." >>>> <kal...@li...> >>>> Message-ID: >>>> >>>> <CY1...@CY...> >>>> >>>> Content-Type: text/plain; charset="utf-8" >>>> >>>> Also, from the practical standpoint, backoff/discounting weights usually >>>> need to be massaged. Otherwise when the grafted LM is small and the main LM >>>> is large, the little model will tend to shoehorn an utterance into itself >>>> rather than let go of it. In my phone number example, everything becomes >>>> digits once the phone number starts. >>>> >>>> -kkm >>>> >>>>> -----Original Message----- >>>>> From: Daniel Povey [mailto:dp...@gm...] >>>>> Sent: 2015-05-21 1204 >>>>> To: Sean True >>>>> Cc: Kirill Katsnelson; Nagendra Goel; Hainan Xu; kaldi- >>>>> us...@li... >>>>> Subject: Re: [Kaldi-users] LM grafting >>>>> >>>>> The general approach is to create an FST for the little language model, >>>>> and then to use fstreplace to replace instances of a particular symbol >>>>> in the top-level language model, with that FST. >>>>> The tricky part is ensuring that the result is determinizable after >>>>> composing with the lexicon. In general our solution is to add special >>>>> disambiguation symbols at the beginning and end of each of the sub- >>>>> FSTs, and of course making sure that the sub-FSTs are themselves >>>>> determinizable. >>>>> Dan >>>>> >>>>> >>>>> On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> >>>>> wrote: >>>>>> That's a subject of some general interest. Is there a discussion of >>>>>> the general approach that was taken somewhere? >>>>>> >>>>>> -- Sean >>>>>> >>>>>> Sean True >>>>>> Semantic Machines >>>>>> >>>>>> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> >>>>> wrote: >>>>>>> >>>>>>> Nagendra Goel has worked on some example scripts for this type of >>>>>>> thing, and with Hainan we were working on trying to get it cleaned >>>>> up >>>>>>> and checked in, but he's going for an internship so it will have to >>>>>>> wait. But Nagendra might be willing to share it with you. >>>>>>> Dan >>>>>>> >>>>>>> >>>>>>> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson >>>>>>> <kir...@sm...> wrote: >>>>>>>> Suppose I have a language model where one token (a "word") is a >>>>>>>> pointer to a whole another LM. This is a practical case when you >>>>>>>> expect an abrupt change in model, a clear example being "my phone >>>>>>>> number is..." and then you'd expect them rattling a string of >>>>>>>> digits. Is there any support in kaldi for this? >>>>>>>> >>>>>>>> Thanks, >>>>>>>> >>>>>>>> -kkm >>>>>>>> >>>>>>>> ------------------------------------------------------------------ >>>>> - >>>>>>>> ----------- One dashboard for servers and applications across >>>>>>>> Physical-Virtual-Cloud Widest out-of-the-box monitoring support >>>>>>>> with 50+ applications Performance metrics, stats and reports that >>>>>>>> give you Actionable Insights Deep dive visibility with transaction >>>>>>>> tracing using APM Insight. >>>>>>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>>>>>>> _______________________________________________ >>>>>>>> Kaldi-users mailing list >>>>>>>> Kal...@li... >>>>>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >>>>>>> >>>>>>> >>>>>>> -------------------------------------------------------------------- >>>>> - >>>>>>> --------- One dashboard for servers and applications across >>>>>>> Physical-Virtual-Cloud Widest out-of-the-box monitoring support with >>>>>>> 50+ applications Performance metrics, stats and reports that give >>>>> you >>>>>>> Actionable Insights Deep dive visibility with transaction tracing >>>>>>> using APM Insight. >>>>>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>>>>>> _______________________________________________ >>>>>>> Kaldi-users mailing list >>>>>>> Kal...@li... >>>>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >>>> >>>> ------------------------------ >>>> >>>> Message: 3 >>>> Date: Thu, 21 May 2015 15:29:54 -0400 >>>> From: Hainan Xu <hai...@gm...> >>>> Subject: Re: [Kaldi-users] LM grafting >>>> To: Daniel Povey <dp...@gm...> >>>> Cc: Sean True <se...@se...>, >>>> "kal...@li..." >>>> <kal...@li...>, Kirill Katsnelson >>>> <kir...@sm...> >>>> Message-ID: >>>> <CALP+BDZvJP-2cZ+fEJEXaMaVWzgy63mtc=J1E...@ma...> >>>> Content-Type: text/plain; charset="utf-8" >>>> >>>> There is a paper in ICASSP 2015 that described some very similar idea: >>>> >>>> Improved recognition of contact names in voice commands >>>> >>>>> On Thu, May 21, 2015 at 3:04 PM, Daniel Povey <dp...@gm...> wrote: >>>>> >>>>> The general approach is to create an FST for the little language >>>>> model, and then to use fstreplace to replace instances of a particular >>>>> symbol in the top-level language model, with that FST. >>>>> The tricky part is ensuring that the result is determinizable after >>>>> composing with the lexicon. In general our solution is to add special >>>>> disambiguation symbols at the beginning and end of each of the >>>>> sub-FSTs, and of course making sure that the sub-FSTs are themselves >>>>> determinizable. >>>>> Dan >>>>> >>>>> >>>>> On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> >>>>> wrote: >>>>>> That's a subject of some general interest. Is there a discussion of >>>>>> the >>>>>> general approach that was taken somewhere? >>>>>> >>>>>> -- Sean >>>>>> >>>>>> Sean True >>>>>> Semantic Machines >>>>>> >>>>>>> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> >>>>>>> wrote: >>>>>>> >>>>>>> Nagendra Goel has worked on some example scripts for this type of >>>>>>> thing, and with Hainan we were working on trying to get it cleaned up >>>>>>> and checked in, but he's going for an internship so it will have to >>>>>>> wait. But Nagendra might be willing to share it with you. >>>>>>> Dan >>>>>>> >>>>>>> >>>>>>> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson >>>>>>> <kir...@sm...> wrote: >>>>>>>> Suppose I have a language model where one token (a "word") is a >>>>> pointer >>>>>>>> to a whole another LM. This is a practical case when you expect an >>>>> abrupt >>>>>>>> change in model, a clear example being "my phone number is..." and >>>>> then >>>>>>>> you'd expect them rattling a string of digits. Is there any support >>>>> in kaldi >>>>>>>> for this? >>>>>>>> >>>>>>>> Thanks, >>>>>>>> >>>>>>>> -kkm >>>>> >>>>> ------------------------------------------------------------------------------ >>>>>>>> One dashboard for servers and applications across >>>>> Physical-Virtual-Cloud >>>>>>>> Widest out-of-the-box monitoring support with 50+ applications >>>>>>>> Performance metrics, stats and reports that give you Actionable >>>>> Insights >>>>>>>> Deep dive visibility with transaction tracing using APM Insight. >>>>>>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>>>>>>> _______________________________________________ >>>>>>>> Kaldi-users mailing list >>>>>>>> Kal...@li... >>>>>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >>>>> >>>>> ------------------------------------------------------------------------------ >>>>>>> One dashboard for servers and applications across >>>>>>> Physical-Virtual-Cloud >>>>>>> Widest out-of-the-box monitoring support with 50+ applications >>>>>>> Performance metrics, stats and reports that give you Actionable >>>>>>> Insights >>>>>>> Deep dive visibility with transaction tracing using APM Insight. >>>>>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>>>>>> _______________________________________________ >>>>>>> Kaldi-users mailing list >>>>>>> Kal...@li... >>>>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >>>> >>>> >>>> >>>> -- >>>> - Hainan >>>> -------------- next part -------------- >>>> An HTML attachment was scrubbed... >>>> >>>> ------------------------------ >>>> >>>> Message: 4 >>>> Date: Thu, 21 May 2015 15:01:51 -0400 >>>> From: Sean True <se...@se...> >>>> Subject: Re: [Kaldi-users] LM grafting >>>> To: Daniel Povey <dp...@gm...> >>>> Cc: Hainan Xu <hai...@gm...>, >>>> "kal...@li..." >>>> <kal...@li...>, Kirill Katsnelson >>>> <kir...@sm...> >>>> Message-ID: >>>> <CALtEaHntdAcmO_Ji5dxsPnT8i9M_LVuGnY0UjkJUPp=pY...@ma...> >>>> Content-Type: text/plain; charset="utf-8" >>>> >>>> That's a subject of some general interest. Is there a discussion of the >>>> general approach that was taken somewhere? >>>> >>>> -- Sean >>>> >>>> Sean True >>>> Semantic Machines >>>> >>>>> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> wrote: >>>>> >>>>> Nagendra Goel has worked on some example scripts for this type of >>>>> thing, and with Hainan we were working on trying to get it cleaned up >>>>> and checked in, but he's going for an internship so it will have to >>>>> wait. But Nagendra might be willing to share it with you. >>>>> Dan >>>>> >>>>> >>>>> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson >>>>> <kir...@sm...> wrote: >>>>>> Suppose I have a language model where one token (a "word") is a >>>>>> pointer >>>>> to a whole another LM. This is a practical case when you expect an >>>>> abrupt >>>>> change in model, a clear example being "my phone number is..." and then >>>>> you'd expect them rattling a string of digits. Is there any support in >>>>> kaldi for this? >>>>>> >>>>>> Thanks, >>>>>> >>>>>> -kkm >>>>> >>>>> ------------------------------------------------------------------------------ >>>>>> One dashboard for servers and applications across >>>>>> Physical-Virtual-Cloud >>>>>> Widest out-of-the-box monitoring support with 50+ applications >>>>>> Performance metrics, stats and reports that give you Actionable >>>>>> Insights >>>>>> Deep dive visibility with transaction tracing using APM Insight. >>>>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>>>>> _______________________________________________ >>>>>> Kaldi-users mailing list >>>>>> Kal...@li... >>>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >>>>> >>>>> >>>>> >>>>> ------------------------------------------------------------------------------ >>>>> One dashboard for servers and applications across >>>>> Physical-Virtual-Cloud >>>>> Widest out-of-the-box monitoring support with 50+ applications >>>>> Performance metrics, stats and reports that give you Actionable >>>>> Insights >>>>> Deep dive visibility with transaction tracing using APM Insight. >>>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>>>> _______________________________________________ >>>>> Kaldi-users mailing list >>>>> Kal...@li... >>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >>>> -------------- next part -------------- >>>> An HTML attachment was scrubbed... >>>> >>>> ------------------------------ >>>> >>>> >>>> ------------------------------------------------------------------------------ >>>> One dashboard for servers and applications across Physical-Virtual-Cloud >>>> Widest out-of-the-box monitoring support with 50+ applications >>>> Performance metrics, stats and reports that give you Actionable Insights >>>> Deep dive visibility with transaction tracing using APM Insight. >>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>>> >>>> ------------------------------ >>>> >>>> _______________________________________________ >>>> Kaldi-users mailing list >>>> Kal...@li... >>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >>>> >>>> >>>> End of Kaldi-users Digest, Vol 29, Issue 15 >>>> ******************************************* >>> >>> >>> ------------------------------------------------------------------------------ >>> One dashboard for servers and applications across Physical-Virtual-Cloud >>> Widest out-of-the-box monitoring support with 50+ applications >>> Performance metrics, stats and reports that give you Actionable Insights >>> Deep dive visibility with transaction tracing using APM Insight. >>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>> _______________________________________________ >>> Kaldi-users mailing list >>> Kal...@li... >>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >> >> >> ------------------------------------------------------------------------------ >> One dashboard for servers and applications across Physical-Virtual-Cloud >> Widest out-of-the-box monitoring support with 50+ applications >> Performance metrics, stats and reports that give you Actionable Insights >> Deep dive visibility with transaction tracing using APM Insight. >> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> _______________________________________________ >> Kaldi-users mailing list >> Kal...@li... >> https://lists.sourceforge.net/lists/listinfo/kaldi-users >> |
|
From: Daniel P. <dp...@gm...> - 2015-05-23 23:06:49
|
One possibility is to use a completely open-source setup, e.g. Voxforge, and forget about the "has a clear advantage" requirement. E.g. target anything that looks like a year, and make a grammar for years. Dan On Fri, May 22, 2015 at 6:32 AM, Nagendra Goel <nag...@go...> wrote: > Since I cannot volunteer my enviornment, do you recommend another > enviornment where this can be prototyped and where you can check in some > class lm recipe that has advantage. > > Nagendra > > Nagendra Kumar Goel > > On May 21, 2015 11:01 PM, "Dimitris Vassos" <dva...@gm...> wrote: >> >> +1 for the class-based LMs. I have also been interested in this >> functionality for some time now, so will be more than happy to try out the >> current implementation, if possible. >> >> Thanks >> Dimitris >> >> > On 22 Μαΐ 2015, at 01:34, kal...@li... >> > wrote: >> > >> > Send Kaldi-users mailing list submissions to >> > kal...@li... >> > >> > To subscribe or unsubscribe via the World Wide Web, visit >> > https://lists.sourceforge.net/lists/listinfo/kaldi-users >> > or, via email, send a message with subject or body 'help' to >> > kal...@li... >> > >> > You can reach the person managing the list at >> > kal...@li... >> > >> > When replying, please edit your Subject line so it is more specific >> > than "Re: Contents of Kaldi-users digest..." >> > >> > >> > Today's Topics: >> > >> > 1. Re: LM grafting (Daniel Povey) >> > 2. Re: LM grafting (Kirill Katsnelson) >> > 3. Re: LM grafting (Hainan Xu) >> > 4. Re: LM grafting (Sean True) >> > >> > >> > ---------------------------------------------------------------------- >> > >> > Message: 1 >> > Date: Thu, 21 May 2015 15:04:04 -0400 >> > From: Daniel Povey <dp...@gm...> >> > Subject: Re: [Kaldi-users] LM grafting >> > To: Sean True <se...@se...> >> > Cc: Hainan Xu <hai...@gm...>, >> > "kal...@li..." >> > <kal...@li...>, Kirill Katsnelson >> > <kir...@sm...> >> > Message-ID: >> > <CAEWAuySHaXwdNJZAoL6CanzHth=k4Y...@ma...> >> > Content-Type: text/plain; charset=UTF-8 >> > >> > The general approach is to create an FST for the little language >> > model, and then to use fstreplace to replace instances of a particular >> > symbol in the top-level language model, with that FST. >> > The tricky part is ensuring that the result is determinizable after >> > composing with the lexicon. In general our solution is to add special >> > disambiguation symbols at the beginning and end of each of the >> > sub-FSTs, and of course making sure that the sub-FSTs are themselves >> > determinizable. >> > Dan >> > >> > >> >> On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> >> >> wrote: >> >> That's a subject of some general interest. Is there a discussion of the >> >> general approach that was taken somewhere? >> >> >> >> -- Sean >> >> >> >> Sean True >> >> Semantic Machines >> >> >> >>> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> >> >>> wrote: >> >>> >> >>> Nagendra Goel has worked on some example scripts for this type of >> >>> thing, and with Hainan we were working on trying to get it cleaned up >> >>> and checked in, but he's going for an internship so it will have to >> >>> wait. But Nagendra might be willing to share it with you. >> >>> Dan >> >>> >> >>> >> >>> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson >> >>> <kir...@sm...> wrote: >> >>>> Suppose I have a language model where one token (a "word") is a >> >>>> pointer >> >>>> to a whole another LM. This is a practical case when you expect an >> >>>> abrupt >> >>>> change in model, a clear example being "my phone number is..." and >> >>>> then >> >>>> you'd expect them rattling a string of digits. Is there any support >> >>>> in kaldi >> >>>> for this? >> >>>> >> >>>> Thanks, >> >>>> >> >>>> -kkm >> >>>> >> >>>> >> >>>> ------------------------------------------------------------------------------ >> >>>> One dashboard for servers and applications across >> >>>> Physical-Virtual-Cloud >> >>>> Widest out-of-the-box monitoring support with 50+ applications >> >>>> Performance metrics, stats and reports that give you Actionable >> >>>> Insights >> >>>> Deep dive visibility with transaction tracing using APM Insight. >> >>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> >>>> _______________________________________________ >> >>>> Kaldi-users mailing list >> >>>> Kal...@li... >> >>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >> >>> >> >>> >> >>> >> >>> ------------------------------------------------------------------------------ >> >>> One dashboard for servers and applications across >> >>> Physical-Virtual-Cloud >> >>> Widest out-of-the-box monitoring support with 50+ applications >> >>> Performance metrics, stats and reports that give you Actionable >> >>> Insights >> >>> Deep dive visibility with transaction tracing using APM Insight. >> >>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> >>> _______________________________________________ >> >>> Kaldi-users mailing list >> >>> Kal...@li... >> >>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >> > >> > >> > >> > ------------------------------ >> > >> > Message: 2 >> > Date: Thu, 21 May 2015 19:24:38 +0000 >> > From: Kirill Katsnelson <kir...@sm...> >> > Subject: Re: [Kaldi-users] LM grafting >> > To: "dp...@gm..." <dp...@gm...>, Sean True >> > <se...@se...> >> > Cc: Hainan Xu <hai...@gm...>, >> > "kal...@li..." >> > <kal...@li...> >> > Message-ID: >> > >> > <CY1...@CY...> >> > >> > Content-Type: text/plain; charset="utf-8" >> > >> > Also, from the practical standpoint, backoff/discounting weights usually >> > need to be massaged. Otherwise when the grafted LM is small and the main LM >> > is large, the little model will tend to shoehorn an utterance into itself >> > rather than let go of it. In my phone number example, everything becomes >> > digits once the phone number starts. >> > >> > -kkm >> > >> >> -----Original Message----- >> >> From: Daniel Povey [mailto:dp...@gm...] >> >> Sent: 2015-05-21 1204 >> >> To: Sean True >> >> Cc: Kirill Katsnelson; Nagendra Goel; Hainan Xu; kaldi- >> >> us...@li... >> >> Subject: Re: [Kaldi-users] LM grafting >> >> >> >> The general approach is to create an FST for the little language model, >> >> and then to use fstreplace to replace instances of a particular symbol >> >> in the top-level language model, with that FST. >> >> The tricky part is ensuring that the result is determinizable after >> >> composing with the lexicon. In general our solution is to add special >> >> disambiguation symbols at the beginning and end of each of the sub- >> >> FSTs, and of course making sure that the sub-FSTs are themselves >> >> determinizable. >> >> Dan >> >> >> >> >> >> On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> >> >> wrote: >> >>> That's a subject of some general interest. Is there a discussion of >> >>> the general approach that was taken somewhere? >> >>> >> >>> -- Sean >> >>> >> >>> Sean True >> >>> Semantic Machines >> >>> >> >>> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> >> >> wrote: >> >>>> >> >>>> Nagendra Goel has worked on some example scripts for this type of >> >>>> thing, and with Hainan we were working on trying to get it cleaned >> >> up >> >>>> and checked in, but he's going for an internship so it will have to >> >>>> wait. But Nagendra might be willing to share it with you. >> >>>> Dan >> >>>> >> >>>> >> >>>> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson >> >>>> <kir...@sm...> wrote: >> >>>>> Suppose I have a language model where one token (a "word") is a >> >>>>> pointer to a whole another LM. This is a practical case when you >> >>>>> expect an abrupt change in model, a clear example being "my phone >> >>>>> number is..." and then you'd expect them rattling a string of >> >>>>> digits. Is there any support in kaldi for this? >> >>>>> >> >>>>> Thanks, >> >>>>> >> >>>>> -kkm >> >>>>> >> >>>>> ------------------------------------------------------------------ >> >> - >> >>>>> ----------- One dashboard for servers and applications across >> >>>>> Physical-Virtual-Cloud Widest out-of-the-box monitoring support >> >>>>> with 50+ applications Performance metrics, stats and reports that >> >>>>> give you Actionable Insights Deep dive visibility with transaction >> >>>>> tracing using APM Insight. >> >>>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> >>>>> _______________________________________________ >> >>>>> Kaldi-users mailing list >> >>>>> Kal...@li... >> >>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >> >>>> >> >>>> >> >>>> -------------------------------------------------------------------- >> >> - >> >>>> --------- One dashboard for servers and applications across >> >>>> Physical-Virtual-Cloud Widest out-of-the-box monitoring support with >> >>>> 50+ applications Performance metrics, stats and reports that give >> >> you >> >>>> Actionable Insights Deep dive visibility with transaction tracing >> >>>> using APM Insight. >> >>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> >>>> _______________________________________________ >> >>>> Kaldi-users mailing list >> >>>> Kal...@li... >> >>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >> > >> > ------------------------------ >> > >> > Message: 3 >> > Date: Thu, 21 May 2015 15:29:54 -0400 >> > From: Hainan Xu <hai...@gm...> >> > Subject: Re: [Kaldi-users] LM grafting >> > To: Daniel Povey <dp...@gm...> >> > Cc: Sean True <se...@se...>, >> > "kal...@li..." >> > <kal...@li...>, Kirill Katsnelson >> > <kir...@sm...> >> > Message-ID: >> > <CALP+BDZvJP-2cZ+fEJEXaMaVWzgy63mtc=J1E...@ma...> >> > Content-Type: text/plain; charset="utf-8" >> > >> > There is a paper in ICASSP 2015 that described some very similar idea: >> > >> > Improved recognition of contact names in voice commands >> > >> >> On Thu, May 21, 2015 at 3:04 PM, Daniel Povey <dp...@gm...> wrote: >> >> >> >> The general approach is to create an FST for the little language >> >> model, and then to use fstreplace to replace instances of a particular >> >> symbol in the top-level language model, with that FST. >> >> The tricky part is ensuring that the result is determinizable after >> >> composing with the lexicon. In general our solution is to add special >> >> disambiguation symbols at the beginning and end of each of the >> >> sub-FSTs, and of course making sure that the sub-FSTs are themselves >> >> determinizable. >> >> Dan >> >> >> >> >> >> On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> >> >> wrote: >> >>> That's a subject of some general interest. Is there a discussion of >> >>> the >> >>> general approach that was taken somewhere? >> >>> >> >>> -- Sean >> >>> >> >>> Sean True >> >>> Semantic Machines >> >>> >> >>>> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> >> >>>> wrote: >> >>>> >> >>>> Nagendra Goel has worked on some example scripts for this type of >> >>>> thing, and with Hainan we were working on trying to get it cleaned up >> >>>> and checked in, but he's going for an internship so it will have to >> >>>> wait. But Nagendra might be willing to share it with you. >> >>>> Dan >> >>>> >> >>>> >> >>>> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson >> >>>> <kir...@sm...> wrote: >> >>>>> Suppose I have a language model where one token (a "word") is a >> >> pointer >> >>>>> to a whole another LM. This is a practical case when you expect an >> >> abrupt >> >>>>> change in model, a clear example being "my phone number is..." and >> >> then >> >>>>> you'd expect them rattling a string of digits. Is there any support >> >> in kaldi >> >>>>> for this? >> >>>>> >> >>>>> Thanks, >> >>>>> >> >>>>> -kkm >> >> >> >> ------------------------------------------------------------------------------ >> >>>>> One dashboard for servers and applications across >> >> Physical-Virtual-Cloud >> >>>>> Widest out-of-the-box monitoring support with 50+ applications >> >>>>> Performance metrics, stats and reports that give you Actionable >> >> Insights >> >>>>> Deep dive visibility with transaction tracing using APM Insight. >> >>>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> >>>>> _______________________________________________ >> >>>>> Kaldi-users mailing list >> >>>>> Kal...@li... >> >>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >> >> >> >> ------------------------------------------------------------------------------ >> >>>> One dashboard for servers and applications across >> >>>> Physical-Virtual-Cloud >> >>>> Widest out-of-the-box monitoring support with 50+ applications >> >>>> Performance metrics, stats and reports that give you Actionable >> >>>> Insights >> >>>> Deep dive visibility with transaction tracing using APM Insight. >> >>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> >>>> _______________________________________________ >> >>>> Kaldi-users mailing list >> >>>> Kal...@li... >> >>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >> > >> > >> > >> > -- >> > - Hainan >> > -------------- next part -------------- >> > An HTML attachment was scrubbed... >> > >> > ------------------------------ >> > >> > Message: 4 >> > Date: Thu, 21 May 2015 15:01:51 -0400 >> > From: Sean True <se...@se...> >> > Subject: Re: [Kaldi-users] LM grafting >> > To: Daniel Povey <dp...@gm...> >> > Cc: Hainan Xu <hai...@gm...>, >> > "kal...@li..." >> > <kal...@li...>, Kirill Katsnelson >> > <kir...@sm...> >> > Message-ID: >> > <CALtEaHntdAcmO_Ji5dxsPnT8i9M_LVuGnY0UjkJUPp=pY...@ma...> >> > Content-Type: text/plain; charset="utf-8" >> > >> > That's a subject of some general interest. Is there a discussion of the >> > general approach that was taken somewhere? >> > >> > -- Sean >> > >> > Sean True >> > Semantic Machines >> > >> >> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> wrote: >> >> >> >> Nagendra Goel has worked on some example scripts for this type of >> >> thing, and with Hainan we were working on trying to get it cleaned up >> >> and checked in, but he's going for an internship so it will have to >> >> wait. But Nagendra might be willing to share it with you. >> >> Dan >> >> >> >> >> >> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson >> >> <kir...@sm...> wrote: >> >>> Suppose I have a language model where one token (a "word") is a >> >>> pointer >> >> to a whole another LM. This is a practical case when you expect an >> >> abrupt >> >> change in model, a clear example being "my phone number is..." and then >> >> you'd expect them rattling a string of digits. Is there any support in >> >> kaldi for this? >> >>> >> >>> Thanks, >> >>> >> >>> -kkm >> >> >> >> ------------------------------------------------------------------------------ >> >>> One dashboard for servers and applications across >> >>> Physical-Virtual-Cloud >> >>> Widest out-of-the-box monitoring support with 50+ applications >> >>> Performance metrics, stats and reports that give you Actionable >> >>> Insights >> >>> Deep dive visibility with transaction tracing using APM Insight. >> >>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> >>> _______________________________________________ >> >>> Kaldi-users mailing list >> >>> Kal...@li... >> >>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >> >> >> >> >> >> >> >> ------------------------------------------------------------------------------ >> >> One dashboard for servers and applications across >> >> Physical-Virtual-Cloud >> >> Widest out-of-the-box monitoring support with 50+ applications >> >> Performance metrics, stats and reports that give you Actionable >> >> Insights >> >> Deep dive visibility with transaction tracing using APM Insight. >> >> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> >> _______________________________________________ >> >> Kaldi-users mailing list >> >> Kal...@li... >> >> https://lists.sourceforge.net/lists/listinfo/kaldi-users >> > -------------- next part -------------- >> > An HTML attachment was scrubbed... >> > >> > ------------------------------ >> > >> > >> > ------------------------------------------------------------------------------ >> > One dashboard for servers and applications across Physical-Virtual-Cloud >> > Widest out-of-the-box monitoring support with 50+ applications >> > Performance metrics, stats and reports that give you Actionable Insights >> > Deep dive visibility with transaction tracing using APM Insight. >> > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> > >> > ------------------------------ >> > >> > _______________________________________________ >> > Kaldi-users mailing list >> > Kal...@li... >> > https://lists.sourceforge.net/lists/listinfo/kaldi-users >> > >> > >> > End of Kaldi-users Digest, Vol 29, Issue 15 >> > ******************************************* >> >> >> ------------------------------------------------------------------------------ >> One dashboard for servers and applications across Physical-Virtual-Cloud >> Widest out-of-the-box monitoring support with 50+ applications >> Performance metrics, stats and reports that give you Actionable Insights >> Deep dive visibility with transaction tracing using APM Insight. >> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> _______________________________________________ >> Kaldi-users mailing list >> Kal...@li... >> https://lists.sourceforge.net/lists/listinfo/kaldi-users > > > ------------------------------------------------------------------------------ > One dashboard for servers and applications across Physical-Virtual-Cloud > Widest out-of-the-box monitoring support with 50+ applications > Performance metrics, stats and reports that give you Actionable Insights > Deep dive visibility with transaction tracing using APM Insight. > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > _______________________________________________ > Kaldi-users mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-users > |
|
From: Nagendra G. <nag...@go...> - 2015-05-22 10:32:55
|
Since I cannot volunteer my enviornment, do you recommend another enviornment where this can be prototyped and where you can check in some class lm recipe that has advantage. Nagendra Nagendra Kumar Goel On May 21, 2015 11:01 PM, "Dimitris Vassos" <dva...@gm...> wrote: > +1 for the class-based LMs. I have also been interested in this > functionality for some time now, so will be more than happy to try out the > current implementation, if possible. > > Thanks > Dimitris > > > On 22 Μαΐ 2015, at 01:34, kal...@li... > wrote: > > > > Send Kaldi-users mailing list submissions to > > kal...@li... > > > > To subscribe or unsubscribe via the World Wide Web, visit > > https://lists.sourceforge.net/lists/listinfo/kaldi-users > > or, via email, send a message with subject or body 'help' to > > kal...@li... > > > > You can reach the person managing the list at > > kal...@li... > > > > When replying, please edit your Subject line so it is more specific > > than "Re: Contents of Kaldi-users digest..." > > > > > > Today's Topics: > > > > 1. Re: LM grafting (Daniel Povey) > > 2. Re: LM grafting (Kirill Katsnelson) > > 3. Re: LM grafting (Hainan Xu) > > 4. Re: LM grafting (Sean True) > > > > > > ---------------------------------------------------------------------- > > > > Message: 1 > > Date: Thu, 21 May 2015 15:04:04 -0400 > > From: Daniel Povey <dp...@gm...> > > Subject: Re: [Kaldi-users] LM grafting > > To: Sean True <se...@se...> > > Cc: Hainan Xu <hai...@gm...>, > > "kal...@li..." > > <kal...@li...>, Kirill Katsnelson > > <kir...@sm...> > > Message-ID: > > <CAEWAuySHaXwdNJZAoL6CanzHth=k4Y...@ma...> > > Content-Type: text/plain; charset=UTF-8 > > > > The general approach is to create an FST for the little language > > model, and then to use fstreplace to replace instances of a particular > > symbol in the top-level language model, with that FST. > > The tricky part is ensuring that the result is determinizable after > > composing with the lexicon. In general our solution is to add special > > disambiguation symbols at the beginning and end of each of the > > sub-FSTs, and of course making sure that the sub-FSTs are themselves > > determinizable. > > Dan > > > > > >> On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> > wrote: > >> That's a subject of some general interest. Is there a discussion of the > >> general approach that was taken somewhere? > >> > >> -- Sean > >> > >> Sean True > >> Semantic Machines > >> > >>> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> > wrote: > >>> > >>> Nagendra Goel has worked on some example scripts for this type of > >>> thing, and with Hainan we were working on trying to get it cleaned up > >>> and checked in, but he's going for an internship so it will have to > >>> wait. But Nagendra might be willing to share it with you. > >>> Dan > >>> > >>> > >>> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson > >>> <kir...@sm...> wrote: > >>>> Suppose I have a language model where one token (a "word") is a > pointer > >>>> to a whole another LM. This is a practical case when you expect an > abrupt > >>>> change in model, a clear example being "my phone number is..." and > then > >>>> you'd expect them rattling a string of digits. Is there any support > in kaldi > >>>> for this? > >>>> > >>>> Thanks, > >>>> > >>>> -kkm > >>>> > >>>> > ------------------------------------------------------------------------------ > >>>> One dashboard for servers and applications across > Physical-Virtual-Cloud > >>>> Widest out-of-the-box monitoring support with 50+ applications > >>>> Performance metrics, stats and reports that give you Actionable > Insights > >>>> Deep dive visibility with transaction tracing using APM Insight. > >>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > >>>> _______________________________________________ > >>>> Kaldi-users mailing list > >>>> Kal...@li... > >>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users > >>> > >>> > >>> > ------------------------------------------------------------------------------ > >>> One dashboard for servers and applications across > Physical-Virtual-Cloud > >>> Widest out-of-the-box monitoring support with 50+ applications > >>> Performance metrics, stats and reports that give you Actionable > Insights > >>> Deep dive visibility with transaction tracing using APM Insight. > >>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > >>> _______________________________________________ > >>> Kaldi-users mailing list > >>> Kal...@li... > >>> https://lists.sourceforge.net/lists/listinfo/kaldi-users > > > > > > > > ------------------------------ > > > > Message: 2 > > Date: Thu, 21 May 2015 19:24:38 +0000 > > From: Kirill Katsnelson <kir...@sm...> > > Subject: Re: [Kaldi-users] LM grafting > > To: "dp...@gm..." <dp...@gm...>, Sean True > > <se...@se...> > > Cc: Hainan Xu <hai...@gm...>, > > "kal...@li..." > > <kal...@li...> > > Message-ID: > > < > CY1...@CY... > > > > > > Content-Type: text/plain; charset="utf-8" > > > > Also, from the practical standpoint, backoff/discounting weights usually > need to be massaged. Otherwise when the grafted LM is small and the main LM > is large, the little model will tend to shoehorn an utterance into itself > rather than let go of it. In my phone number example, everything becomes > digits once the phone number starts. > > > > -kkm > > > >> -----Original Message----- > >> From: Daniel Povey [mailto:dp...@gm...] > >> Sent: 2015-05-21 1204 > >> To: Sean True > >> Cc: Kirill Katsnelson; Nagendra Goel; Hainan Xu; kaldi- > >> us...@li... > >> Subject: Re: [Kaldi-users] LM grafting > >> > >> The general approach is to create an FST for the little language model, > >> and then to use fstreplace to replace instances of a particular symbol > >> in the top-level language model, with that FST. > >> The tricky part is ensuring that the result is determinizable after > >> composing with the lexicon. In general our solution is to add special > >> disambiguation symbols at the beginning and end of each of the sub- > >> FSTs, and of course making sure that the sub-FSTs are themselves > >> determinizable. > >> Dan > >> > >> > >> On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> > >> wrote: > >>> That's a subject of some general interest. Is there a discussion of > >>> the general approach that was taken somewhere? > >>> > >>> -- Sean > >>> > >>> Sean True > >>> Semantic Machines > >>> > >>> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> > >> wrote: > >>>> > >>>> Nagendra Goel has worked on some example scripts for this type of > >>>> thing, and with Hainan we were working on trying to get it cleaned > >> up > >>>> and checked in, but he's going for an internship so it will have to > >>>> wait. But Nagendra might be willing to share it with you. > >>>> Dan > >>>> > >>>> > >>>> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson > >>>> <kir...@sm...> wrote: > >>>>> Suppose I have a language model where one token (a "word") is a > >>>>> pointer to a whole another LM. This is a practical case when you > >>>>> expect an abrupt change in model, a clear example being "my phone > >>>>> number is..." and then you'd expect them rattling a string of > >>>>> digits. Is there any support in kaldi for this? > >>>>> > >>>>> Thanks, > >>>>> > >>>>> -kkm > >>>>> > >>>>> ------------------------------------------------------------------ > >> - > >>>>> ----------- One dashboard for servers and applications across > >>>>> Physical-Virtual-Cloud Widest out-of-the-box monitoring support > >>>>> with 50+ applications Performance metrics, stats and reports that > >>>>> give you Actionable Insights Deep dive visibility with transaction > >>>>> tracing using APM Insight. > >>>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > >>>>> _______________________________________________ > >>>>> Kaldi-users mailing list > >>>>> Kal...@li... > >>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users > >>>> > >>>> > >>>> -------------------------------------------------------------------- > >> - > >>>> --------- One dashboard for servers and applications across > >>>> Physical-Virtual-Cloud Widest out-of-the-box monitoring support with > >>>> 50+ applications Performance metrics, stats and reports that give > >> you > >>>> Actionable Insights Deep dive visibility with transaction tracing > >>>> using APM Insight. > >>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > >>>> _______________________________________________ > >>>> Kaldi-users mailing list > >>>> Kal...@li... > >>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users > > > > ------------------------------ > > > > Message: 3 > > Date: Thu, 21 May 2015 15:29:54 -0400 > > From: Hainan Xu <hai...@gm...> > > Subject: Re: [Kaldi-users] LM grafting > > To: Daniel Povey <dp...@gm...> > > Cc: Sean True <se...@se...>, > > "kal...@li..." > > <kal...@li...>, Kirill Katsnelson > > <kir...@sm...> > > Message-ID: > > <CALP+BDZvJP-2cZ+fEJEXaMaVWzgy63mtc=J1E...@ma...> > > Content-Type: text/plain; charset="utf-8" > > > > There is a paper in ICASSP 2015 that described some very similar idea: > > > > Improved recognition of contact names in voice commands > > > >> On Thu, May 21, 2015 at 3:04 PM, Daniel Povey <dp...@gm...> wrote: > >> > >> The general approach is to create an FST for the little language > >> model, and then to use fstreplace to replace instances of a particular > >> symbol in the top-level language model, with that FST. > >> The tricky part is ensuring that the result is determinizable after > >> composing with the lexicon. In general our solution is to add special > >> disambiguation symbols at the beginning and end of each of the > >> sub-FSTs, and of course making sure that the sub-FSTs are themselves > >> determinizable. > >> Dan > >> > >> > >> On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> > >> wrote: > >>> That's a subject of some general interest. Is there a discussion of the > >>> general approach that was taken somewhere? > >>> > >>> -- Sean > >>> > >>> Sean True > >>> Semantic Machines > >>> > >>>> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> > wrote: > >>>> > >>>> Nagendra Goel has worked on some example scripts for this type of > >>>> thing, and with Hainan we were working on trying to get it cleaned up > >>>> and checked in, but he's going for an internship so it will have to > >>>> wait. But Nagendra might be willing to share it with you. > >>>> Dan > >>>> > >>>> > >>>> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson > >>>> <kir...@sm...> wrote: > >>>>> Suppose I have a language model where one token (a "word") is a > >> pointer > >>>>> to a whole another LM. This is a practical case when you expect an > >> abrupt > >>>>> change in model, a clear example being "my phone number is..." and > >> then > >>>>> you'd expect them rattling a string of digits. Is there any support > >> in kaldi > >>>>> for this? > >>>>> > >>>>> Thanks, > >>>>> > >>>>> -kkm > >> > ------------------------------------------------------------------------------ > >>>>> One dashboard for servers and applications across > >> Physical-Virtual-Cloud > >>>>> Widest out-of-the-box monitoring support with 50+ applications > >>>>> Performance metrics, stats and reports that give you Actionable > >> Insights > >>>>> Deep dive visibility with transaction tracing using APM Insight. > >>>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > >>>>> _______________________________________________ > >>>>> Kaldi-users mailing list > >>>>> Kal...@li... > >>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users > >> > ------------------------------------------------------------------------------ > >>>> One dashboard for servers and applications across > Physical-Virtual-Cloud > >>>> Widest out-of-the-box monitoring support with 50+ applications > >>>> Performance metrics, stats and reports that give you Actionable > Insights > >>>> Deep dive visibility with transaction tracing using APM Insight. > >>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > >>>> _______________________________________________ > >>>> Kaldi-users mailing list > >>>> Kal...@li... > >>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users > > > > > > > > -- > > - Hainan > > -------------- next part -------------- > > An HTML attachment was scrubbed... > > > > ------------------------------ > > > > Message: 4 > > Date: Thu, 21 May 2015 15:01:51 -0400 > > From: Sean True <se...@se...> > > Subject: Re: [Kaldi-users] LM grafting > > To: Daniel Povey <dp...@gm...> > > Cc: Hainan Xu <hai...@gm...>, > > "kal...@li..." > > <kal...@li...>, Kirill Katsnelson > > <kir...@sm...> > > Message-ID: > > <CALtEaHntdAcmO_Ji5dxsPnT8i9M_LVuGnY0UjkJUPp=pY...@ma...> > > Content-Type: text/plain; charset="utf-8" > > > > That's a subject of some general interest. Is there a discussion of the > > general approach that was taken somewhere? > > > > -- Sean > > > > Sean True > > Semantic Machines > > > >> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> wrote: > >> > >> Nagendra Goel has worked on some example scripts for this type of > >> thing, and with Hainan we were working on trying to get it cleaned up > >> and checked in, but he's going for an internship so it will have to > >> wait. But Nagendra might be willing to share it with you. > >> Dan > >> > >> > >> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson > >> <kir...@sm...> wrote: > >>> Suppose I have a language model where one token (a "word") is a pointer > >> to a whole another LM. This is a practical case when you expect an > abrupt > >> change in model, a clear example being "my phone number is..." and then > >> you'd expect them rattling a string of digits. Is there any support in > >> kaldi for this? > >>> > >>> Thanks, > >>> > >>> -kkm > >> > ------------------------------------------------------------------------------ > >>> One dashboard for servers and applications across > Physical-Virtual-Cloud > >>> Widest out-of-the-box monitoring support with 50+ applications > >>> Performance metrics, stats and reports that give you Actionable > Insights > >>> Deep dive visibility with transaction tracing using APM Insight. > >>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > >>> _______________________________________________ > >>> Kaldi-users mailing list > >>> Kal...@li... > >>> https://lists.sourceforge.net/lists/listinfo/kaldi-users > >> > >> > >> > ------------------------------------------------------------------------------ > >> One dashboard for servers and applications across Physical-Virtual-Cloud > >> Widest out-of-the-box monitoring support with 50+ applications > >> Performance metrics, stats and reports that give you Actionable Insights > >> Deep dive visibility with transaction tracing using APM Insight. > >> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > >> _______________________________________________ > >> Kaldi-users mailing list > >> Kal...@li... > >> https://lists.sourceforge.net/lists/listinfo/kaldi-users > > -------------- next part -------------- > > An HTML attachment was scrubbed... > > > > ------------------------------ > > > > > ------------------------------------------------------------------------------ > > One dashboard for servers and applications across Physical-Virtual-Cloud > > Widest out-of-the-box monitoring support with 50+ applications > > Performance metrics, stats and reports that give you Actionable Insights > > Deep dive visibility with transaction tracing using APM Insight. > > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > > > > ------------------------------ > > > > _______________________________________________ > > Kaldi-users mailing list > > Kal...@li... > > https://lists.sourceforge.net/lists/listinfo/kaldi-users > > > > > > End of Kaldi-users Digest, Vol 29, Issue 15 > > ******************************************* > > > ------------------------------------------------------------------------------ > One dashboard for servers and applications across Physical-Virtual-Cloud > Widest out-of-the-box monitoring support with 50+ applications > Performance metrics, stats and reports that give you Actionable Insights > Deep dive visibility with transaction tracing using APM Insight. > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > _______________________________________________ > Kaldi-users mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-users > |
|
From: Dimitris V. <dva...@gm...> - 2015-05-22 03:00:18
|
+1 for the class-based LMs. I have also been interested in this functionality for some time now, so will be more than happy to try out the current implementation, if possible. Thanks Dimitris > On 22 Μαΐ 2015, at 01:34, kal...@li... wrote: > > Send Kaldi-users mailing list submissions to > kal...@li... > > To subscribe or unsubscribe via the World Wide Web, visit > https://lists.sourceforge.net/lists/listinfo/kaldi-users > or, via email, send a message with subject or body 'help' to > kal...@li... > > You can reach the person managing the list at > kal...@li... > > When replying, please edit your Subject line so it is more specific > than "Re: Contents of Kaldi-users digest..." > > > Today's Topics: > > 1. Re: LM grafting (Daniel Povey) > 2. Re: LM grafting (Kirill Katsnelson) > 3. Re: LM grafting (Hainan Xu) > 4. Re: LM grafting (Sean True) > > > ---------------------------------------------------------------------- > > Message: 1 > Date: Thu, 21 May 2015 15:04:04 -0400 > From: Daniel Povey <dp...@gm...> > Subject: Re: [Kaldi-users] LM grafting > To: Sean True <se...@se...> > Cc: Hainan Xu <hai...@gm...>, > "kal...@li..." > <kal...@li...>, Kirill Katsnelson > <kir...@sm...> > Message-ID: > <CAEWAuySHaXwdNJZAoL6CanzHth=k4Y...@ma...> > Content-Type: text/plain; charset=UTF-8 > > The general approach is to create an FST for the little language > model, and then to use fstreplace to replace instances of a particular > symbol in the top-level language model, with that FST. > The tricky part is ensuring that the result is determinizable after > composing with the lexicon. In general our solution is to add special > disambiguation symbols at the beginning and end of each of the > sub-FSTs, and of course making sure that the sub-FSTs are themselves > determinizable. > Dan > > >> On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> wrote: >> That's a subject of some general interest. Is there a discussion of the >> general approach that was taken somewhere? >> >> -- Sean >> >> Sean True >> Semantic Machines >> >>> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> wrote: >>> >>> Nagendra Goel has worked on some example scripts for this type of >>> thing, and with Hainan we were working on trying to get it cleaned up >>> and checked in, but he's going for an internship so it will have to >>> wait. But Nagendra might be willing to share it with you. >>> Dan >>> >>> >>> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson >>> <kir...@sm...> wrote: >>>> Suppose I have a language model where one token (a "word") is a pointer >>>> to a whole another LM. This is a practical case when you expect an abrupt >>>> change in model, a clear example being "my phone number is..." and then >>>> you'd expect them rattling a string of digits. Is there any support in kaldi >>>> for this? >>>> >>>> Thanks, >>>> >>>> -kkm >>>> >>>> ------------------------------------------------------------------------------ >>>> One dashboard for servers and applications across Physical-Virtual-Cloud >>>> Widest out-of-the-box monitoring support with 50+ applications >>>> Performance metrics, stats and reports that give you Actionable Insights >>>> Deep dive visibility with transaction tracing using APM Insight. >>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>>> _______________________________________________ >>>> Kaldi-users mailing list >>>> Kal...@li... >>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >>> >>> >>> ------------------------------------------------------------------------------ >>> One dashboard for servers and applications across Physical-Virtual-Cloud >>> Widest out-of-the-box monitoring support with 50+ applications >>> Performance metrics, stats and reports that give you Actionable Insights >>> Deep dive visibility with transaction tracing using APM Insight. >>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>> _______________________________________________ >>> Kaldi-users mailing list >>> Kal...@li... >>> https://lists.sourceforge.net/lists/listinfo/kaldi-users > > > > ------------------------------ > > Message: 2 > Date: Thu, 21 May 2015 19:24:38 +0000 > From: Kirill Katsnelson <kir...@sm...> > Subject: Re: [Kaldi-users] LM grafting > To: "dp...@gm..." <dp...@gm...>, Sean True > <se...@se...> > Cc: Hainan Xu <hai...@gm...>, > "kal...@li..." > <kal...@li...> > Message-ID: > <CY1...@CY...> > > Content-Type: text/plain; charset="utf-8" > > Also, from the practical standpoint, backoff/discounting weights usually need to be massaged. Otherwise when the grafted LM is small and the main LM is large, the little model will tend to shoehorn an utterance into itself rather than let go of it. In my phone number example, everything becomes digits once the phone number starts. > > -kkm > >> -----Original Message----- >> From: Daniel Povey [mailto:dp...@gm...] >> Sent: 2015-05-21 1204 >> To: Sean True >> Cc: Kirill Katsnelson; Nagendra Goel; Hainan Xu; kaldi- >> us...@li... >> Subject: Re: [Kaldi-users] LM grafting >> >> The general approach is to create an FST for the little language model, >> and then to use fstreplace to replace instances of a particular symbol >> in the top-level language model, with that FST. >> The tricky part is ensuring that the result is determinizable after >> composing with the lexicon. In general our solution is to add special >> disambiguation symbols at the beginning and end of each of the sub- >> FSTs, and of course making sure that the sub-FSTs are themselves >> determinizable. >> Dan >> >> >> On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> >> wrote: >>> That's a subject of some general interest. Is there a discussion of >>> the general approach that was taken somewhere? >>> >>> -- Sean >>> >>> Sean True >>> Semantic Machines >>> >>> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> >> wrote: >>>> >>>> Nagendra Goel has worked on some example scripts for this type of >>>> thing, and with Hainan we were working on trying to get it cleaned >> up >>>> and checked in, but he's going for an internship so it will have to >>>> wait. But Nagendra might be willing to share it with you. >>>> Dan >>>> >>>> >>>> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson >>>> <kir...@sm...> wrote: >>>>> Suppose I have a language model where one token (a "word") is a >>>>> pointer to a whole another LM. This is a practical case when you >>>>> expect an abrupt change in model, a clear example being "my phone >>>>> number is..." and then you'd expect them rattling a string of >>>>> digits. Is there any support in kaldi for this? >>>>> >>>>> Thanks, >>>>> >>>>> -kkm >>>>> >>>>> ------------------------------------------------------------------ >> - >>>>> ----------- One dashboard for servers and applications across >>>>> Physical-Virtual-Cloud Widest out-of-the-box monitoring support >>>>> with 50+ applications Performance metrics, stats and reports that >>>>> give you Actionable Insights Deep dive visibility with transaction >>>>> tracing using APM Insight. >>>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>>>> _______________________________________________ >>>>> Kaldi-users mailing list >>>>> Kal...@li... >>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >>>> >>>> >>>> -------------------------------------------------------------------- >> - >>>> --------- One dashboard for servers and applications across >>>> Physical-Virtual-Cloud Widest out-of-the-box monitoring support with >>>> 50+ applications Performance metrics, stats and reports that give >> you >>>> Actionable Insights Deep dive visibility with transaction tracing >>>> using APM Insight. >>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>>> _______________________________________________ >>>> Kaldi-users mailing list >>>> Kal...@li... >>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users > > ------------------------------ > > Message: 3 > Date: Thu, 21 May 2015 15:29:54 -0400 > From: Hainan Xu <hai...@gm...> > Subject: Re: [Kaldi-users] LM grafting > To: Daniel Povey <dp...@gm...> > Cc: Sean True <se...@se...>, > "kal...@li..." > <kal...@li...>, Kirill Katsnelson > <kir...@sm...> > Message-ID: > <CALP+BDZvJP-2cZ+fEJEXaMaVWzgy63mtc=J1E...@ma...> > Content-Type: text/plain; charset="utf-8" > > There is a paper in ICASSP 2015 that described some very similar idea: > > Improved recognition of contact names in voice commands > >> On Thu, May 21, 2015 at 3:04 PM, Daniel Povey <dp...@gm...> wrote: >> >> The general approach is to create an FST for the little language >> model, and then to use fstreplace to replace instances of a particular >> symbol in the top-level language model, with that FST. >> The tricky part is ensuring that the result is determinizable after >> composing with the lexicon. In general our solution is to add special >> disambiguation symbols at the beginning and end of each of the >> sub-FSTs, and of course making sure that the sub-FSTs are themselves >> determinizable. >> Dan >> >> >> On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> >> wrote: >>> That's a subject of some general interest. Is there a discussion of the >>> general approach that was taken somewhere? >>> >>> -- Sean >>> >>> Sean True >>> Semantic Machines >>> >>>> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> wrote: >>>> >>>> Nagendra Goel has worked on some example scripts for this type of >>>> thing, and with Hainan we were working on trying to get it cleaned up >>>> and checked in, but he's going for an internship so it will have to >>>> wait. But Nagendra might be willing to share it with you. >>>> Dan >>>> >>>> >>>> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson >>>> <kir...@sm...> wrote: >>>>> Suppose I have a language model where one token (a "word") is a >> pointer >>>>> to a whole another LM. This is a practical case when you expect an >> abrupt >>>>> change in model, a clear example being "my phone number is..." and >> then >>>>> you'd expect them rattling a string of digits. Is there any support >> in kaldi >>>>> for this? >>>>> >>>>> Thanks, >>>>> >>>>> -kkm >> ------------------------------------------------------------------------------ >>>>> One dashboard for servers and applications across >> Physical-Virtual-Cloud >>>>> Widest out-of-the-box monitoring support with 50+ applications >>>>> Performance metrics, stats and reports that give you Actionable >> Insights >>>>> Deep dive visibility with transaction tracing using APM Insight. >>>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>>>> _______________________________________________ >>>>> Kaldi-users mailing list >>>>> Kal...@li... >>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >> ------------------------------------------------------------------------------ >>>> One dashboard for servers and applications across Physical-Virtual-Cloud >>>> Widest out-of-the-box monitoring support with 50+ applications >>>> Performance metrics, stats and reports that give you Actionable Insights >>>> Deep dive visibility with transaction tracing using APM Insight. >>>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>>> _______________________________________________ >>>> Kaldi-users mailing list >>>> Kal...@li... >>>> https://lists.sourceforge.net/lists/listinfo/kaldi-users > > > > -- > - Hainan > -------------- next part -------------- > An HTML attachment was scrubbed... > > ------------------------------ > > Message: 4 > Date: Thu, 21 May 2015 15:01:51 -0400 > From: Sean True <se...@se...> > Subject: Re: [Kaldi-users] LM grafting > To: Daniel Povey <dp...@gm...> > Cc: Hainan Xu <hai...@gm...>, > "kal...@li..." > <kal...@li...>, Kirill Katsnelson > <kir...@sm...> > Message-ID: > <CALtEaHntdAcmO_Ji5dxsPnT8i9M_LVuGnY0UjkJUPp=pY...@ma...> > Content-Type: text/plain; charset="utf-8" > > That's a subject of some general interest. Is there a discussion of the > general approach that was taken somewhere? > > -- Sean > > Sean True > Semantic Machines > >> On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> wrote: >> >> Nagendra Goel has worked on some example scripts for this type of >> thing, and with Hainan we were working on trying to get it cleaned up >> and checked in, but he's going for an internship so it will have to >> wait. But Nagendra might be willing to share it with you. >> Dan >> >> >> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson >> <kir...@sm...> wrote: >>> Suppose I have a language model where one token (a "word") is a pointer >> to a whole another LM. This is a practical case when you expect an abrupt >> change in model, a clear example being "my phone number is..." and then >> you'd expect them rattling a string of digits. Is there any support in >> kaldi for this? >>> >>> Thanks, >>> >>> -kkm >> ------------------------------------------------------------------------------ >>> One dashboard for servers and applications across Physical-Virtual-Cloud >>> Widest out-of-the-box monitoring support with 50+ applications >>> Performance metrics, stats and reports that give you Actionable Insights >>> Deep dive visibility with transaction tracing using APM Insight. >>> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >>> _______________________________________________ >>> Kaldi-users mailing list >>> Kal...@li... >>> https://lists.sourceforge.net/lists/listinfo/kaldi-users >> >> >> ------------------------------------------------------------------------------ >> One dashboard for servers and applications across Physical-Virtual-Cloud >> Widest out-of-the-box monitoring support with 50+ applications >> Performance metrics, stats and reports that give you Actionable Insights >> Deep dive visibility with transaction tracing using APM Insight. >> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> _______________________________________________ >> Kaldi-users mailing list >> Kal...@li... >> https://lists.sourceforge.net/lists/listinfo/kaldi-users > -------------- next part -------------- > An HTML attachment was scrubbed... > > ------------------------------ > > ------------------------------------------------------------------------------ > One dashboard for servers and applications across Physical-Virtual-Cloud > Widest out-of-the-box monitoring support with 50+ applications > Performance metrics, stats and reports that give you Actionable Insights > Deep dive visibility with transaction tracing using APM Insight. > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > > ------------------------------ > > _______________________________________________ > Kaldi-users mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-users > > > End of Kaldi-users Digest, Vol 29, Issue 15 > ******************************************* |
|
From: Sean T. <se...@se...> - 2015-05-21 19:33:08
|
That's a subject of some general interest. Is there a discussion of the general approach that was taken somewhere? -- Sean Sean True Semantic Machines On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> wrote: > Nagendra Goel has worked on some example scripts for this type of > thing, and with Hainan we were working on trying to get it cleaned up > and checked in, but he's going for an internship so it will have to > wait. But Nagendra might be willing to share it with you. > Dan > > > On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson > <kir...@sm...> wrote: > > Suppose I have a language model where one token (a "word") is a pointer > to a whole another LM. This is a practical case when you expect an abrupt > change in model, a clear example being "my phone number is..." and then > you'd expect them rattling a string of digits. Is there any support in > kaldi for this? > > > > Thanks, > > > > -kkm > > > ------------------------------------------------------------------------------ > > One dashboard for servers and applications across Physical-Virtual-Cloud > > Widest out-of-the-box monitoring support with 50+ applications > > Performance metrics, stats and reports that give you Actionable Insights > > Deep dive visibility with transaction tracing using APM Insight. > > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > > _______________________________________________ > > Kaldi-users mailing list > > Kal...@li... > > https://lists.sourceforge.net/lists/listinfo/kaldi-users > > > ------------------------------------------------------------------------------ > One dashboard for servers and applications across Physical-Virtual-Cloud > Widest out-of-the-box monitoring support with 50+ applications > Performance metrics, stats and reports that give you Actionable Insights > Deep dive visibility with transaction tracing using APM Insight. > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > _______________________________________________ > Kaldi-users mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-users > |
|
From: Hainan Xu <hai...@gm...> - 2015-05-21 19:30:01
|
There is a paper in ICASSP 2015 that described some very similar idea: Improved recognition of contact names in voice commands On Thu, May 21, 2015 at 3:04 PM, Daniel Povey <dp...@gm...> wrote: > The general approach is to create an FST for the little language > model, and then to use fstreplace to replace instances of a particular > symbol in the top-level language model, with that FST. > The tricky part is ensuring that the result is determinizable after > composing with the lexicon. In general our solution is to add special > disambiguation symbols at the beginning and end of each of the > sub-FSTs, and of course making sure that the sub-FSTs are themselves > determinizable. > Dan > > > On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> > wrote: > > That's a subject of some general interest. Is there a discussion of the > > general approach that was taken somewhere? > > > > -- Sean > > > > Sean True > > Semantic Machines > > > > On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> wrote: > >> > >> Nagendra Goel has worked on some example scripts for this type of > >> thing, and with Hainan we were working on trying to get it cleaned up > >> and checked in, but he's going for an internship so it will have to > >> wait. But Nagendra might be willing to share it with you. > >> Dan > >> > >> > >> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson > >> <kir...@sm...> wrote: > >> > Suppose I have a language model where one token (a "word") is a > pointer > >> > to a whole another LM. This is a practical case when you expect an > abrupt > >> > change in model, a clear example being "my phone number is..." and > then > >> > you'd expect them rattling a string of digits. Is there any support > in kaldi > >> > for this? > >> > > >> > Thanks, > >> > > >> > -kkm > >> > > >> > > ------------------------------------------------------------------------------ > >> > One dashboard for servers and applications across > Physical-Virtual-Cloud > >> > Widest out-of-the-box monitoring support with 50+ applications > >> > Performance metrics, stats and reports that give you Actionable > Insights > >> > Deep dive visibility with transaction tracing using APM Insight. > >> > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > >> > _______________________________________________ > >> > Kaldi-users mailing list > >> > Kal...@li... > >> > https://lists.sourceforge.net/lists/listinfo/kaldi-users > >> > >> > >> > ------------------------------------------------------------------------------ > >> One dashboard for servers and applications across Physical-Virtual-Cloud > >> Widest out-of-the-box monitoring support with 50+ applications > >> Performance metrics, stats and reports that give you Actionable Insights > >> Deep dive visibility with transaction tracing using APM Insight. > >> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > >> _______________________________________________ > >> Kaldi-users mailing list > >> Kal...@li... > >> https://lists.sourceforge.net/lists/listinfo/kaldi-users > > > > > -- - Hainan |
|
From: Kirill K. <kir...@sm...> - 2015-05-21 19:24:51
|
Also, from the practical standpoint, backoff/discounting weights usually need to be massaged. Otherwise when the grafted LM is small and the main LM is large, the little model will tend to shoehorn an utterance into itself rather than let go of it. In my phone number example, everything becomes digits once the phone number starts. -kkm > -----Original Message----- > From: Daniel Povey [mailto:dp...@gm...] > Sent: 2015-05-21 1204 > To: Sean True > Cc: Kirill Katsnelson; Nagendra Goel; Hainan Xu; kaldi- > us...@li... > Subject: Re: [Kaldi-users] LM grafting > > The general approach is to create an FST for the little language model, > and then to use fstreplace to replace instances of a particular symbol > in the top-level language model, with that FST. > The tricky part is ensuring that the result is determinizable after > composing with the lexicon. In general our solution is to add special > disambiguation symbols at the beginning and end of each of the sub- > FSTs, and of course making sure that the sub-FSTs are themselves > determinizable. > Dan > > > On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> > wrote: > > That's a subject of some general interest. Is there a discussion of > > the general approach that was taken somewhere? > > > > -- Sean > > > > Sean True > > Semantic Machines > > > > On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> > wrote: > >> > >> Nagendra Goel has worked on some example scripts for this type of > >> thing, and with Hainan we were working on trying to get it cleaned > up > >> and checked in, but he's going for an internship so it will have to > >> wait. But Nagendra might be willing to share it with you. > >> Dan > >> > >> > >> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson > >> <kir...@sm...> wrote: > >> > Suppose I have a language model where one token (a "word") is a > >> > pointer to a whole another LM. This is a practical case when you > >> > expect an abrupt change in model, a clear example being "my phone > >> > number is..." and then you'd expect them rattling a string of > >> > digits. Is there any support in kaldi for this? > >> > > >> > Thanks, > >> > > >> > -kkm > >> > > >> > ------------------------------------------------------------------ > - > >> > ----------- One dashboard for servers and applications across > >> > Physical-Virtual-Cloud Widest out-of-the-box monitoring support > >> > with 50+ applications Performance metrics, stats and reports that > >> > give you Actionable Insights Deep dive visibility with transaction > >> > tracing using APM Insight. > >> > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > >> > _______________________________________________ > >> > Kaldi-users mailing list > >> > Kal...@li... > >> > https://lists.sourceforge.net/lists/listinfo/kaldi-users > >> > >> > >> -------------------------------------------------------------------- > - > >> --------- One dashboard for servers and applications across > >> Physical-Virtual-Cloud Widest out-of-the-box monitoring support with > >> 50+ applications Performance metrics, stats and reports that give > you > >> Actionable Insights Deep dive visibility with transaction tracing > >> using APM Insight. > >> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > >> _______________________________________________ > >> Kaldi-users mailing list > >> Kal...@li... > >> https://lists.sourceforge.net/lists/listinfo/kaldi-users > > > > |
|
From: Daniel P. <dp...@gm...> - 2015-05-21 19:04:10
|
The general approach is to create an FST for the little language model, and then to use fstreplace to replace instances of a particular symbol in the top-level language model, with that FST. The tricky part is ensuring that the result is determinizable after composing with the lexicon. In general our solution is to add special disambiguation symbols at the beginning and end of each of the sub-FSTs, and of course making sure that the sub-FSTs are themselves determinizable. Dan On Thu, May 21, 2015 at 3:01 PM, Sean True <se...@se...> wrote: > That's a subject of some general interest. Is there a discussion of the > general approach that was taken somewhere? > > -- Sean > > Sean True > Semantic Machines > > On Thu, May 21, 2015 at 2:14 PM, Daniel Povey <dp...@gm...> wrote: >> >> Nagendra Goel has worked on some example scripts for this type of >> thing, and with Hainan we were working on trying to get it cleaned up >> and checked in, but he's going for an internship so it will have to >> wait. But Nagendra might be willing to share it with you. >> Dan >> >> >> On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson >> <kir...@sm...> wrote: >> > Suppose I have a language model where one token (a "word") is a pointer >> > to a whole another LM. This is a practical case when you expect an abrupt >> > change in model, a clear example being "my phone number is..." and then >> > you'd expect them rattling a string of digits. Is there any support in kaldi >> > for this? >> > >> > Thanks, >> > >> > -kkm >> > >> > ------------------------------------------------------------------------------ >> > One dashboard for servers and applications across Physical-Virtual-Cloud >> > Widest out-of-the-box monitoring support with 50+ applications >> > Performance metrics, stats and reports that give you Actionable Insights >> > Deep dive visibility with transaction tracing using APM Insight. >> > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> > _______________________________________________ >> > Kaldi-users mailing list >> > Kal...@li... >> > https://lists.sourceforge.net/lists/listinfo/kaldi-users >> >> >> ------------------------------------------------------------------------------ >> One dashboard for servers and applications across Physical-Virtual-Cloud >> Widest out-of-the-box monitoring support with 50+ applications >> Performance metrics, stats and reports that give you Actionable Insights >> Deep dive visibility with transaction tracing using APM Insight. >> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> _______________________________________________ >> Kaldi-users mailing list >> Kal...@li... >> https://lists.sourceforge.net/lists/listinfo/kaldi-users > > |
|
From: Daniel P. <dp...@gm...> - 2015-05-21 18:14:51
|
Nagendra Goel has worked on some example scripts for this type of thing, and with Hainan we were working on trying to get it cleaned up and checked in, but he's going for an internship so it will have to wait. But Nagendra might be willing to share it with you. Dan On Thu, May 21, 2015 at 2:10 PM, Kirill Katsnelson <kir...@sm...> wrote: > Suppose I have a language model where one token (a "word") is a pointer to a whole another LM. This is a practical case when you expect an abrupt change in model, a clear example being "my phone number is..." and then you'd expect them rattling a string of digits. Is there any support in kaldi for this? > > Thanks, > > -kkm > ------------------------------------------------------------------------------ > One dashboard for servers and applications across Physical-Virtual-Cloud > Widest out-of-the-box monitoring support with 50+ applications > Performance metrics, stats and reports that give you Actionable Insights > Deep dive visibility with transaction tracing using APM Insight. > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > _______________________________________________ > Kaldi-users mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-users |
|
From: Kirill K. <kir...@sm...> - 2015-05-21 18:10:33
|
Suppose I have a language model where one token (a "word") is a pointer to a whole another LM. This is a practical case when you expect an abrupt change in model, a clear example being "my phone number is..." and then you'd expect them rattling a string of digits. Is there any support in kaldi for this? Thanks, -kkm |
|
From: Xingyu Na <asr...@gm...> - 2015-05-21 02:48:35
|
Right.... Thanks! X. On 05/21/2015 10:45 AM, Vijayaditya Peddinti wrote: > > You could use a combination of splice component and FixedAffine > component to achieve this. > > Vijay > > On May 20, 2015 10:39 PM, "Xingyu Na" <asr...@gm... > <mailto:asr...@gm...>> wrote: > > Hi, > > I want to apply a non-rectangle window to the spliced feature. > I checked the splice-feats code and seems that it is not > supported, does it? > > X. > > ------------------------------------------------------------------------------ > One dashboard for servers and applications across > Physical-Virtual-Cloud > Widest out-of-the-box monitoring support with 50+ applications > Performance metrics, stats and reports that give you Actionable > Insights > Deep dive visibility with transaction tracing using APM Insight. > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > _______________________________________________ > Kaldi-users mailing list > Kal...@li... > <mailto:Kal...@li...> > https://lists.sourceforge.net/lists/listinfo/kaldi-users > |
|
From: Vijayaditya P. <p.v...@gm...> - 2015-05-21 02:45:56
|
You could use a combination of splice component and FixedAffine component to achieve this. Vijay On May 20, 2015 10:39 PM, "Xingyu Na" <asr...@gm...> wrote: > Hi, > > I want to apply a non-rectangle window to the spliced feature. > I checked the splice-feats code and seems that it is not supported, does > it? > > X. > > > ------------------------------------------------------------------------------ > One dashboard for servers and applications across Physical-Virtual-Cloud > Widest out-of-the-box monitoring support with 50+ applications > Performance metrics, stats and reports that give you Actionable Insights > Deep dive visibility with transaction tracing using APM Insight. > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > _______________________________________________ > Kaldi-users mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-users > |
|
From: Xingyu Na <asr...@gm...> - 2015-05-21 02:39:16
|
Hi, I want to apply a non-rectangle window to the spliced feature. I checked the splice-feats code and seems that it is not supported, does it? X. |
|
From: Daniel P. <dp...@gm...> - 2015-05-20 18:47:26
|
Yes, you will. Actually that might be easier, come to think of it. Dan On Wed, May 20, 2015 at 2:19 PM, Kirill Katsnelson <kir...@sm...> wrote: > And if I give apply-cmvn options --norm-means=false --norm-vars=false, will I get exactly same result? > >> -----Original Message----- >> From: Daniel Povey [mailto:dp...@gm...] >> Sent: 2015-05-19 1651 >> To: Kirill Katsnelson >> Cc: kal...@li... >> Subject: Re: [Kaldi-users] Dummy CMVN stats >> >> Yes it's possible, look at the --fake option to >> steps/compute_cmvn_stats.sh Dan >> >> >> On Tue, May 19, 2015 at 7:40 PM, Kirill Katsnelson >> <kir...@sm...> wrote: >> > I want to bypass the cmvn stats application, but it is not easy in >> some existing scripts. Is there a tool to create dummy stats, which, >> when applied, do not change features? The stats is just a vector, but I >> do not know what the elements mean. >> > >> > Thanks, >> > >> > -kkm >> > --------------------------------------------------------------------- >> - >> > -------- One dashboard for servers and applications across >> > Physical-Virtual-Cloud Widest out-of-the-box monitoring support with >> > 50+ applications Performance metrics, stats and reports that give you >> > Actionable Insights Deep dive visibility with transaction tracing >> > using APM Insight. >> > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> > _______________________________________________ >> > Kaldi-users mailing list >> > Kal...@li... >> > https://lists.sourceforge.net/lists/listinfo/kaldi-users |
|
From: Kirill K. <kir...@sm...> - 2015-05-20 18:42:19
|
Need to pick a collective brain on a strange issue. My test set comes from a different source than the training set. The former is telephone recorded utterances, and the former is TEDlium, downsampled and filtered through an IIR approximation of an average phone line response. Different sources usually mean that CM[V]N will be beneficial. However, I am seeing a statistically significant 2% absolute *surge* in WER in a set trained with CMN (no variance normalization) versus a set with no normalization. The model is essentially Karel's DNN from the TEDlium recipe, with lda features, but no fMLLR. Has anyone seen anything like that? I am looking if not for an explanation, then at the least for an anecdote that this has happened to someone before. I'm still trying to exclude a pilot error, but not finding it yet. Thanks, -kkm |
|
From: Kirill K. <kir...@sm...> - 2015-05-20 18:19:19
|
And if I give apply-cmvn options --norm-means=false --norm-vars=false, will I get exactly same result? > -----Original Message----- > From: Daniel Povey [mailto:dp...@gm...] > Sent: 2015-05-19 1651 > To: Kirill Katsnelson > Cc: kal...@li... > Subject: Re: [Kaldi-users] Dummy CMVN stats > > Yes it's possible, look at the --fake option to > steps/compute_cmvn_stats.sh Dan > > > On Tue, May 19, 2015 at 7:40 PM, Kirill Katsnelson > <kir...@sm...> wrote: > > I want to bypass the cmvn stats application, but it is not easy in > some existing scripts. Is there a tool to create dummy stats, which, > when applied, do not change features? The stats is just a vector, but I > do not know what the elements mean. > > > > Thanks, > > > > -kkm > > --------------------------------------------------------------------- > - > > -------- One dashboard for servers and applications across > > Physical-Virtual-Cloud Widest out-of-the-box monitoring support with > > 50+ applications Performance metrics, stats and reports that give you > > Actionable Insights Deep dive visibility with transaction tracing > > using APM Insight. > > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > > _______________________________________________ > > Kaldi-users mailing list > > Kal...@li... > > https://lists.sourceforge.net/lists/listinfo/kaldi-users |
|
From: Kirill K. <kir...@sm...> - 2015-05-20 00:02:45
|
> -----Original Message----- > From: Daniel Povey [mailto:dp...@gm...] > Sent: 2015-05-19 1651 > > Yes it's possible, look at the --fake option to > steps/compute_cmvn_stats.sh Dan This is it, thanks! -kkm > On Tue, May 19, 2015 at 7:40 PM, Kirill Katsnelson > <kir...@sm...> wrote: > > I want to bypass the cmvn stats application, but it is not easy in > some existing scripts. Is there a tool to create dummy stats, which, > when applied, do not change features? The stats is just a vector, but I > do not know what the elements mean. > > > > Thanks, > > > > -kkm > > --------------------------------------------------------------------- > - > > -------- One dashboard for servers and applications across > > Physical-Virtual-Cloud Widest out-of-the-box monitoring support with > > 50+ applications Performance metrics, stats and reports that give you > > Actionable Insights Deep dive visibility with transaction tracing > > using APM Insight. > > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > > _______________________________________________ > > Kaldi-users mailing list > > Kal...@li... > > https://lists.sourceforge.net/lists/listinfo/kaldi-users |
|
From: Daniel P. <dp...@gm...> - 2015-05-19 23:51:35
|
Yes it's possible, look at the --fake option to steps/compute_cmvn_stats.sh Dan On Tue, May 19, 2015 at 7:40 PM, Kirill Katsnelson <kir...@sm...> wrote: > I want to bypass the cmvn stats application, but it is not easy in some existing scripts. Is there a tool to create dummy stats, which, when applied, do not change features? The stats is just a vector, but I do not know what the elements mean. > > Thanks, > > -kkm > ------------------------------------------------------------------------------ > One dashboard for servers and applications across Physical-Virtual-Cloud > Widest out-of-the-box monitoring support with 50+ applications > Performance metrics, stats and reports that give you Actionable Insights > Deep dive visibility with transaction tracing using APM Insight. > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > _______________________________________________ > Kaldi-users mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-users |
|
From: Kirill K. <kir...@sm...> - 2015-05-19 23:40:31
|
I want to bypass the cmvn stats application, but it is not easy in some existing scripts. Is there a tool to create dummy stats, which, when applied, do not change features? The stats is just a vector, but I do not know what the elements mean. Thanks, -kkm |
|
From: Daniel P. <dp...@gm...> - 2015-05-19 18:29:43
|
You haven't given enough information for us to tell what the error was. The part of the log that you showed only had normal warnings in it, nothing that would cause it to fail. And it's not clear to me that the log which you showed was the one that had an error (only one out of the 4 failed). Please try to figure it out yourself for a while before following up with the list, though. This is probably a very simple problem that can be remedied just by looking at the error message in the log file. Dan On Tue, May 19, 2015 at 2:20 PM, Naresh kumar <ell...@gm...> wrote: > Dear all, > I am trying to build word level ASR for English Database. MFCC features are > extracted. While running monophone training I found the following error. > > steps/train_mono.sh: Compiling training graphs > steps/train_mono.sh: Aligning data equally (pass 0) > run.pl: 1 / 4 failed, log is in exp/mono/log/align.0.*.log > > The exp/mono/log/align.0.1.log file looks like > > WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features > for utterance P28_S46 > WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features > for utterance P28_S47 > WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features > for utterance P28_S48 > WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features > for utterance P28_S49 > WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features > for utterance P28_S5 > WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features > for utterance P28_S50 > WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features > for utterance P28_S6 > WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features > for utterance P28_S7 > WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features > for utterance P28_S8 > WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features > for utterance P28_S9 > WARNING (align-equal-compiled:main():align-equal-compiled.cc:109) Computed > 640 alignments; 10 lacked features, 0 had other errors. > WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for > utterance P10_W1-10_1 > WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for > utterance P10_W10-10_10 > WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for > utterance P10_W100-10_100 > WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for > utterance P10_W101-10_101 > WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for > utterance P10_W102-10_102 > WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for > utterance P10_W103-10_103 > WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for > utterance P10_W104-10_104 > WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for > utterance P10_W105-10_105 > > Please let me know how to solve the issue > > -- > > > Regards > Naresh Kumar > > ------------------------------------------------------------------------------ > One dashboard for servers and applications across Physical-Virtual-Cloud > Widest out-of-the-box monitoring support with 50+ applications > Performance metrics, stats and reports that give you Actionable Insights > Deep dive visibility with transaction tracing using APM Insight. > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > _______________________________________________ > Kaldi-users mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-users > |
|
From: Naresh k. <ell...@gm...> - 2015-05-19 18:21:04
|
Dear all, I am trying to build word level ASR for English Database. MFCC features are extracted. While running monophone training I found the following error. *steps/train_mono.sh: Compiling training graphssteps/train_mono.sh: Aligning data equally (pass 0)run.pl <http://run.pl>: 1 / 4 failed, log is in exp/mono/log/align.0.*.log* The exp/mono/log/align.0.1.log file looks like *WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features for utterance P28_S46WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features for utterance P28_S47WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features for utterance P28_S48WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features for utterance P28_S49WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features for utterance P28_S5WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features for utterance P28_S50WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features for utterance P28_S6WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features for utterance P28_S7WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features for utterance P28_S8WARNING (align-equal-compiled:main():align-equal-compiled.cc:69) No features for utterance P28_S9WARNING (align-equal-compiled:main():align-equal-compiled.cc:109) Computed 640 alignments; 10 lacked features, 0 had other errors.WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for utterance P10_W1-10_1WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for utterance P10_W10-10_10WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for utterance P10_W100-10_100WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for utterance P10_W101-10_101WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for utterance P10_W102-10_102WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for utterance P10_W103-10_103WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for utterance P10_W104-10_104WARNING (gmm-acc-stats-ali:main():gmm-acc-stats-ali.cc:79) No alignment for utterance P10_W105-10_105* Please let me know how to solve the issue -- Regards Naresh Kumar |
|
From: Daniel P. <dp...@gm...> - 2015-05-18 19:48:10
|
> > My targets are (phone,HMM-state) extracted from an alignments' file using > TransitionIdToPhone() and TransitionIdToHmmState(). Is this the correct way > to get them? If yes, why are they not affected by the decision tree, since > ,as far as I understand, TransitionIds and alignments are? Yes, that is the correct way to get them; but they are not affected by the decision tree, i.e. you will get the same answer regardless of the decision tree. It would be much better to use pdf-id as the target for neural net training; this is what is standard in DNN-based speech recognition, and it's what Kaldi uses. Dan > > Thanks a lot for your time and help, > > Yannis > >> >> It's a little unclear from your question what you problem is. If your >> targets are really (phone, HMM-state) [e.g. phone = 1 ... 40, >> hmm-state = 1 ... 3], then the decision tree is not involved at all. >> If your targets are really pairs (phone, pdf-id) in Kaldi terminology, >> then yes, some of your targets might be quite infrequent (especially >> if, in your system, there are word-position-dependent phones). If the >> issue is caused by word-position-dependent phones, then thresholds >> like you suggested won't help- in that case it would be better simply >> to use pdf-id as the target. Otherwise (if there are no >> word-position-dependent phones), then applying a state-count threshold >> could help with your immediate problem of low counts, but I'm quite >> doubtful that it would make any difference to WER in the end. >> One thing that can be helpful in dealing with these very low-count >> states is, when dividing by the prior probabilities of states, to >> divide by the average state posterior computed over the training data, >> instead of the prior of that state in your training labels. This >> stops a particular pathology where particular low-count states can get >> too likely after dividing by the prior. >> >> Dan >> >> >> >> > >> >> >> >> >> >> The reason why that wasn't implemented as a feature is that due to the >> >> way the tree-clustering was implemented, it would be slightly tricky >> >> to implement. It would be possible but would require attention from >> >> someone familiar with the tree-clustering code. (e.g. Hainan >> >> Do you have any definite reason to believe that this would improve >> >> results? >> >> Dan >> >> >> >> >> >> On Sun, May 17, 2015 at 4:58 AM, Yannis Chalkiadakis >> >> <ha...@ho...> wrote: >> >> > Hi all, >> >> > >> >> > I would like to prevent a leaf from splitting when its size is less >> >> > than >> >> > , >> >> > say , N, or merge it with its parent during bottom-up clustering if >> >> > its >> >> > size >> >> > is less than N. Is this possible? How could I extract the size of the >> >> > leaf? >> >> > >> >> > Thank you in advance for your time and help, >> >> > Yannis >> >> > >> >> > >> >> > >> >> > >> >> > >> >> > >> >> > ------------------------------------------------------------------------------ >> >> > One dashboard for servers and applications across >> >> > Physical-Virtual-Cloud >> >> > Widest out-of-the-box monitoring support with 50+ applications >> >> > Performance metrics, stats and reports that give you Actionable >> >> > Insights >> >> > Deep dive visibility with transaction tracing using APM Insight. >> >> > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> >> > _______________________________________________ >> >> > Kaldi-users mailing list >> >> > Kal...@li... >> >> > https://lists.sourceforge.net/lists/listinfo/kaldi-users >> >> > |
|
From: Nagendra G. <nag...@go...> - 2015-05-18 10:12:16
|
Matthew, If you are already working on decision trees, I am wondering if you could allow for full-covariance models on the distance calculations (interesting in some cases ). Also if you are using MDL, allow for correlation between frames by just scaling down the number of frames by a user defined constant. Nagendra Kumar Goel On May 18, 2015 3:27 AM, "Matthew Aylett" <mat...@gm...> wrote: > Hi > > This is peripheral but a good point to describe the extensions of tree > building we have been putting in to kaldi for speech synthesis. > > I think this is currently within the sandbox/idlak branch (but could be > merged across). > > 1. We allow full context models which can be generated by the idlak front > end. > > This extends the phone model name to a vector of integers where each > integer represents a context (such as previous syllable stressed, or > current word part of speech etc. Currently this sits a little uncomfortably > with the current scheme that generates a vector based on phone context > length (i.e. 5 for a quinphone), and uses an integer representing phone + > word position. However it is compatible with this. > > 2. We allow the addition of manual question sets to the tree building > system. i.e Is subsequent syllable stressed? > > This is something required for speech synthesis (although semi automated), > and not something ASR is very keen on I believe as it adds an arbitrary > manual stage. > > 3. I'm planning to add an option to use minimum description length as a > element in the tree building process. This would (I think) lead to deeper > trees. > > These changes a primarily to model prosodic effects (and I realise there > is a not a great history of prosody being much use for ASR), but I thought > it would be useful to bring these changes to kaldi community's attention in > case ASR was interested in any of them. > > v best > > Matthew > > > On Sun, May 17, 2015 at 11:27 PM, Yannis Chalkiadakis <ha...@ho... > > wrote: > >> Hi Dan, >> >> Thanks for the prompt reply. >> >> What I want to do is use pairs of (phone,HMM-state) as the targets to >> train a DNN. However, after getting these targets I noticed that some of >> them occur rarely in the dataset, e.g 40 times in ~30M vectors of data. >> Consequently, I would like to prune the decision tree (or stop splitting >> during building it) so that eventually all my targets appear at least, say, >> 1000 times each. >> I tried to use a threshold for the likelihood to control splitting, but I >> still did not get the results I expected. >> >> Is there another way to deal with this matter? >> >> Thank you for the help, >> Yannis >> >> > >> > >> > The reason why that wasn't implemented as a feature is that due to the >> > way the tree-clustering was implemented, it would be slightly tricky >> > to implement. It would be possible but would require attention from >> > someone familiar with the tree-clustering code. (e.g. Hainan >> > Do you have any definite reason to believe that this would improve >> results? >> > Dan >> > >> > >> > On Sun, May 17, 2015 at 4:58 AM, Yannis Chalkiadakis >> > <ha...@ho...> wrote: >> > > Hi all, >> > > >> > > I would like to prevent a leaf from splitting when its size is less >> than , >> > > say , N, or merge it with its parent during bottom-up clustering if >> its size >> > > is less than N. Is this possible? How could I extract the size of the >> leaf? >> > > >> > > Thank you in advance for your time and help, >> > > Yannis >> > > >> > > >> > > >> > > >> > > >> ------------------------------------------------------------------------------ >> > > One dashboard for servers and applications across >> Physical-Virtual-Cloud >> > > Widest out-of-the-box monitoring support with 50+ applications >> > > Performance metrics, stats and reports that give you Actionable >> Insights >> > > Deep dive visibility with transaction tracing using APM Insight. >> > > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> > > _______________________________________________ >> > > Kaldi-users mailing list >> > > Kal...@li... >> > > https://lists.sourceforge.net/lists/listinfo/kaldi-users >> > > >> >> >> ------------------------------------------------------------------------------ >> One dashboard for servers and applications across Physical-Virtual-Cloud >> Widest out-of-the-box monitoring support with 50+ applications >> Performance metrics, stats and reports that give you Actionable Insights >> Deep dive visibility with transaction tracing using APM Insight. >> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y >> _______________________________________________ >> Kaldi-users mailing list >> Kal...@li... >> https://lists.sourceforge.net/lists/listinfo/kaldi-users >> >> > > > ------------------------------------------------------------------------------ > One dashboard for servers and applications across Physical-Virtual-Cloud > Widest out-of-the-box monitoring support with 50+ applications > Performance metrics, stats and reports that give you Actionable Insights > Deep dive visibility with transaction tracing using APM Insight. > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > _______________________________________________ > Kaldi-users mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-users > > |
|
From: Yannis C. <ha...@ho...> - 2015-05-18 09:46:22
|
Hi Dan, My targets are (phone,HMM-state) extracted from an alignments' file using TransitionIdToPhone() and TransitionIdToHmmState(). Is this the correct way to get them? If yes, why are they not affected by the decision tree, since ,as far as I understand, TransitionIds and alignments are? Thanks a lot for your time and help, Yannis > > It's a little unclear from your question what you problem is. If your > targets are really (phone, HMM-state) [e.g. phone = 1 ... 40, > hmm-state = 1 ... 3], then the decision tree is not involved at all. > If your targets are really pairs (phone, pdf-id) in Kaldi terminology, > then yes, some of your targets might be quite infrequent (especially > if, in your system, there are word-position-dependent phones). If the > issue is caused by word-position-dependent phones, then thresholds > like you suggested won't help- in that case it would be better simply > to use pdf-id as the target. Otherwise (if there are no > word-position-dependent phones), then applying a state-count threshold > could help with your immediate problem of low counts, but I'm quite > doubtful that it would make any difference to WER in the end. > One thing that can be helpful in dealing with these very low-count > states is, when dividing by the prior probabilities of states, to > divide by the average state posterior computed over the training data, > instead of the prior of that state in your training labels. This > stops a particular pathology where particular low-count states can get > too likely after dividing by the prior. > > Dan > > > > > > >> > >> > >> The reason why that wasn't implemented as a feature is that due to the > >> way the tree-clustering was implemented, it would be slightly tricky > >> to implement. It would be possible but would require attention from > >> someone familiar with the tree-clustering code. (e.g. Hainan > >> Do you have any definite reason to believe that this would improve > >> results? > >> Dan > >> > >> > >> On Sun, May 17, 2015 at 4:58 AM, Yannis Chalkiadakis > >> <ha...@ho...> wrote: > >> > Hi all, > >> > > >> > I would like to prevent a leaf from splitting when its size is less than > >> > , > >> > say , N, or merge it with its parent during bottom-up clustering if its > >> > size > >> > is less than N. Is this possible? How could I extract the size of the > >> > leaf? > >> > > >> > Thank you in advance for your time and help, > >> > Yannis > >> > > >> > > >> > > >> > > >> > > >> > ------------------------------------------------------------------------------ > >> > One dashboard for servers and applications across Physical-Virtual-Cloud > >> > Widest out-of-the-box monitoring support with 50+ applications > >> > Performance metrics, stats and reports that give you Actionable Insights > >> > Deep dive visibility with transaction tracing using APM Insight. > >> > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > >> > _______________________________________________ > >> > Kaldi-users mailing list > >> > Kal...@li... > >> > https://lists.sourceforge.net/lists/listinfo/kaldi-users > >> > |
|
From: Matthew A. <mat...@gm...> - 2015-05-18 07:26:47
|
Hi This is peripheral but a good point to describe the extensions of tree building we have been putting in to kaldi for speech synthesis. I think this is currently within the sandbox/idlak branch (but could be merged across). 1. We allow full context models which can be generated by the idlak front end. This extends the phone model name to a vector of integers where each integer represents a context (such as previous syllable stressed, or current word part of speech etc. Currently this sits a little uncomfortably with the current scheme that generates a vector based on phone context length (i.e. 5 for a quinphone), and uses an integer representing phone + word position. However it is compatible with this. 2. We allow the addition of manual question sets to the tree building system. i.e Is subsequent syllable stressed? This is something required for speech synthesis (although semi automated), and not something ASR is very keen on I believe as it adds an arbitrary manual stage. 3. I'm planning to add an option to use minimum description length as a element in the tree building process. This would (I think) lead to deeper trees. These changes a primarily to model prosodic effects (and I realise there is a not a great history of prosody being much use for ASR), but I thought it would be useful to bring these changes to kaldi community's attention in case ASR was interested in any of them. v best Matthew On Sun, May 17, 2015 at 11:27 PM, Yannis Chalkiadakis <ha...@ho...> wrote: > Hi Dan, > > Thanks for the prompt reply. > > What I want to do is use pairs of (phone,HMM-state) as the targets to > train a DNN. However, after getting these targets I noticed that some of > them occur rarely in the dataset, e.g 40 times in ~30M vectors of data. > Consequently, I would like to prune the decision tree (or stop splitting > during building it) so that eventually all my targets appear at least, say, > 1000 times each. > I tried to use a threshold for the likelihood to control splitting, but I > still did not get the results I expected. > > Is there another way to deal with this matter? > > Thank you for the help, > Yannis > > > > > > > The reason why that wasn't implemented as a feature is that due to the > > way the tree-clustering was implemented, it would be slightly tricky > > to implement. It would be possible but would require attention from > > someone familiar with the tree-clustering code. (e.g. Hainan > > Do you have any definite reason to believe that this would improve > results? > > Dan > > > > > > On Sun, May 17, 2015 at 4:58 AM, Yannis Chalkiadakis > > <ha...@ho...> wrote: > > > Hi all, > > > > > > I would like to prevent a leaf from splitting when its size is less > than , > > > say , N, or merge it with its parent during bottom-up clustering if > its size > > > is less than N. Is this possible? How could I extract the size of the > leaf? > > > > > > Thank you in advance for your time and help, > > > Yannis > > > > > > > > > > > > > > > > ------------------------------------------------------------------------------ > > > One dashboard for servers and applications across > Physical-Virtual-Cloud > > > Widest out-of-the-box monitoring support with 50+ applications > > > Performance metrics, stats and reports that give you Actionable > Insights > > > Deep dive visibility with transaction tracing using APM Insight. > > > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > > > _______________________________________________ > > > Kaldi-users mailing list > > > Kal...@li... > > > https://lists.sourceforge.net/lists/listinfo/kaldi-users > > > > > > ------------------------------------------------------------------------------ > One dashboard for servers and applications across Physical-Virtual-Cloud > Widest out-of-the-box monitoring support with 50+ applications > Performance metrics, stats and reports that give you Actionable Insights > Deep dive visibility with transaction tracing using APM Insight. > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y > _______________________________________________ > Kaldi-users mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-users > > |
1 message has been excluded from this view by a project administrator.
