Menu
▾
▴
kaldi-developers — Kaldi Developers
You can subscribe to this list here.
| 2011 |
Jan
|
Feb
|
Mar
|
Apr
|
May
|
Jun
(4) |
Jul
|
Aug
|
Sep
(1) |
Oct
(4) |
Nov
(1) |
Dec
(14) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2012 |
Jan
(1) |
Feb
(8) |
Mar
|
Apr
(1) |
May
(3) |
Jun
(13) |
Jul
(7) |
Aug
(11) |
Sep
(6) |
Oct
(14) |
Nov
(16) |
Dec
(1) |
| 2013 |
Jan
(3) |
Feb
(8) |
Mar
(17) |
Apr
(21) |
May
(27) |
Jun
(11) |
Jul
(11) |
Aug
(21) |
Sep
(39) |
Oct
(17) |
Nov
(39) |
Dec
(28) |
| 2014 |
Jan
(36) |
Feb
(30) |
Mar
(35) |
Apr
(17) |
May
(22) |
Jun
(28) |
Jul
(23) |
Aug
(41) |
Sep
(17) |
Oct
(10) |
Nov
(22) |
Dec
(56) |
| 2015 |
Jan
(30) |
Feb
(32) |
Mar
(37) |
Apr
(28) |
May
(79) |
Jun
(18) |
Jul
(35) |
Aug
|
Sep
(1) |
Oct
|
Nov
|
Dec
|
|
From: Daniel P. <dp...@gm...> - 2014-06-24 17:08:56
|
The program ali-to-phones should give you the information that you need (when given the right options), but it won't be in a CTM-like format. Dan On Tue, Jun 24, 2014 at 1:06 PM, Arif Khan <ife...@gm...> wrote: > Hi, > > In the "steps/get_train_ctm.sh", we can get the word alignment which looks > something like this: > > 018c021d 1 2.32 0.08 AN > 018c021d 1 2.40 0.46 APPARENT > 018c021d 1 2.86 0.59 SLOWDOWN > 018c021d 1 3.45 0.11 IN > 018c021d 1 3.56 0.58 ECONOMIC > 018c021d 1 4.14 0.30 GROWTH > > Is there any recipe to get the phone alignment. i.e Beginning and End > timing of phones in the same format as above. > > > Best regards, > Arif > > > ------------------------------------------------------------------------------ > Open source business process management suite built on Java and Eclipse > Turn processes into business applications with Bonita BPM Community Edition > Quickly connect people, data, and systems into organized workflows > Winner of BOSSIE, CODIE, OW2 and Gartner awards > http://p.sf.net/sfu/Bonitasoft > _______________________________________________ > Kaldi-developers mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-developers > > |
|
From: Arif K. <ife...@gm...> - 2014-06-24 17:06:48
|
Hi, In the "steps/get_train_ctm.sh", we can get the word alignment which looks something like this: 018c021d 1 2.32 0.08 AN 018c021d 1 2.40 0.46 APPARENT 018c021d 1 2.86 0.59 SLOWDOWN 018c021d 1 3.45 0.11 IN 018c021d 1 3.56 0.58 ECONOMIC 018c021d 1 4.14 0.30 GROWTH Is there any recipe to get the phone alignment. i.e Beginning and End timing of phones in the same format as above. Best regards, Arif |
|
From: Mike D. <md...@gm...> - 2014-06-19 21:28:59
|
With the nnet classes directly accessing parameters is easy (using GetWeights). With nnet2, it is not so straightforward. The right way appears to be using the Vectorize method. Unfortunately, Vectorize is only available for AffineComponent and BlockAffineComponent, not the rest of the Affine component classes. Is this intentional? What is the right way to directly access weights and biases? Thanks! --Mike |
|
From: Daniel P. <dp...@gm...> - 2014-06-19 16:54:20
|
The problem is likely with data/lang/words.txt. It should look like this: # head ~/kaldi-clean/egs/rm/s5/data/lang/words.txt <eps> 0 !SIL 1 A 2 A42128 3 AAW 4 ABERDEEN 5 ABOARD 6 ABOVE 7 ADD 8 ADDED 9 Dan On Mon, Jun 16, 2014 at 5:36 PM, Rohit Kandurwar <roh...@gm...> wrote: > Hi, > > I am trying to run the test scripts of Kaldi on RM dataset. > > However while running the rm_prepare_grammer.sh script I'm facing an error: > > Symbol "AAW" is not mapped to any integer arc ilabel, symbol table = > data/lang/words.txt, source = data/local/tmp/G.txt, line = 1. > > Please help me understand what could be the problem, I am new to speech > processing. > > > Regards > Rohit > > > ------------------------------------------------------------------------------ > HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions > Find What Matters Most in Your Big Data with HPCC Systems > Open Source. Fast. Scalable. Simple. Ideal for Dirty Data. > Leverages Graph Analysis for Fast Processing & Easy Data Exploration > http://p.sf.net/sfu/hpccsystems > _______________________________________________ > Kaldi-developers mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-developers > > |
|
From: Rohit K. <roh...@gm...> - 2014-06-16 21:36:42
|
Hi, I am trying to run the test scripts of Kaldi on RM dataset. However while running the rm_prepare_grammer.sh script I'm facing an error: Symbol "AAW" is not mapped to any integer arc ilabel, symbol table = data/lang/words.txt, source = data/local/tmp/G.txt, line = 1. Please help me understand what could be the problem, I am new to speech processing. Regards Rohit |
|
From: Xavier A. <xan...@gm...> - 2014-06-13 16:27:16
|
great, thanks. X. On Fri, Jun 13, 2014 at 6:21 PM, Daniel Povey <dp...@gm...> wrote: > Yes it did disappear (it related to the determinization but it wasn't > really necessary; the --max-mem is sufficient). You can just ignore it. > Dan > > > > On Fri, Jun 13, 2014 at 7:38 AM, Xavier Anguera <xan...@gm...> > wrote: > >> Hi, >> did the command line parameter --max-arcs disappear from >> the gmm-latgen-faster executable? >> Is there a replacement variable for it? my script is currently setting it >> to -1, should I then consider that it is fine to ignore it? >> >> thanks, >> >> X. Anguera >> >> >> ------------------------------------------------------------------------------ >> HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions >> Find What Matters Most in Your Big Data with HPCC Systems >> Open Source. Fast. Scalable. Simple. Ideal for Dirty Data. >> Leverages Graph Analysis for Fast Processing & Easy Data Exploration >> http://p.sf.net/sfu/hpccsystems >> _______________________________________________ >> Kaldi-developers mailing list >> Kal...@li... >> https://lists.sourceforge.net/lists/listinfo/kaldi-developers >> >> > |
|
From: Daniel P. <dp...@gm...> - 2014-06-13 16:21:59
|
Yes it did disappear (it related to the determinization but it wasn't really necessary; the --max-mem is sufficient). You can just ignore it. Dan On Fri, Jun 13, 2014 at 7:38 AM, Xavier Anguera <xan...@gm...> wrote: > Hi, > did the command line parameter --max-arcs disappear from > the gmm-latgen-faster executable? > Is there a replacement variable for it? my script is currently setting it > to -1, should I then consider that it is fine to ignore it? > > thanks, > > X. Anguera > > > ------------------------------------------------------------------------------ > HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions > Find What Matters Most in Your Big Data with HPCC Systems > Open Source. Fast. Scalable. Simple. Ideal for Dirty Data. > Leverages Graph Analysis for Fast Processing & Easy Data Exploration > http://p.sf.net/sfu/hpccsystems > _______________________________________________ > Kaldi-developers mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-developers > > |
|
From: Xavier A. <xan...@gm...> - 2014-06-13 11:38:58
|
Hi, did the command line parameter --max-arcs disappear from the gmm-latgen-faster executable? Is there a replacement variable for it? my script is currently setting it to -1, should I then consider that it is fine to ignore it? thanks, X. Anguera |
|
From: <jen...@a2...> - 2014-06-11 01:19:30
|
Kaldi - Build # 556 - Failure: See the build log in attachment for the details. |
|
From: Daniel P. <dp...@gm...> - 2014-06-10 18:08:29
|
I recently added a script egs/wsj/s5/local/run_gender_dep.sh which demonstrates MAP adaptation of models. Dan On Tue, Jun 10, 2014 at 8:28 AM, Xavier Anguera <xan...@gm...> wrote: > Hi, > could someone point me to any recipe or code example to perfom acoustic > model adaptation in Kaldi? > I have an acoustic model trained in one acoustic condition and would like > to adapt it (not retrain, I do not have the original data) to another > domain given some acoustic data on that domain. > Standard MAP or MLLR would work for me, although I do not have any speaker > info on the new domain. > > Thanks a lot, > > X. > > > > ------------------------------------------------------------------------------ > HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions > Find What Matters Most in Your Big Data with HPCC Systems > Open Source. Fast. Scalable. Simple. Ideal for Dirty Data. > Leverages Graph Analysis for Fast Processing & Easy Data Exploration > http://p.sf.net/sfu/hpccsystems > _______________________________________________ > Kaldi-developers mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-developers > > |
|
From: Xavier A. <xan...@gm...> - 2014-06-10 12:28:22
|
Hi, could someone point me to any recipe or code example to perfom acoustic model adaptation in Kaldi? I have an acoustic model trained in one acoustic condition and would like to adapt it (not retrain, I do not have the original data) to another domain given some acoustic data on that domain. Standard MAP or MLLR would work for me, although I do not have any speaker info on the new domain. Thanks a lot, X. |
|
From: <jen...@a2...> - 2014-06-09 08:12:51
|
Kaldi - Build # 555 - Fixed: See the build log in attachment for the details. |
|
From: Christopher K. <chr...@a2...> - 2014-06-09 08:09:59
|
Hi Dan, It's again a problem with the coverage estimation, not kaldi compilation. I try to fix it. -- CK On 06/09/2014 07:04 AM, Daniel Povey wrote: > Christopher, could you please get a stack trace for the > logistic-regression-test failure? > I can't replicate it. > Dan > > > > On Mon, Jun 9, 2014 at 12:56 AM, <jen...@a2... > <mailto:jen...@a2...>> wrote: > > Kaldi - Build # 554 - Still Failing: > > See the build log in attachment for the details. > > > ------------------------------------------------------------------------------ > HPCC Systems Open Source Big Data Platform from LexisNexis Risk > Solutions > Find What Matters Most in Your Big Data with HPCC Systems > Open Source. Fast. Scalable. Simple. Ideal for Dirty Data. > Leverages Graph Analysis for Fast Processing & Easy Data Exploration > http://www.hpccsystems.com > _______________________________________________ > Kaldi-developers mailing list > Kal...@li... > <mailto:Kal...@li...> > https://lists.sourceforge.net/lists/listinfo/kaldi-developers > > -- Christopher Kermorvant R&D Manager - A2iA - France 39 rue de la Bienfaisance - 75008 Paris +33 (0) 144 420 086 / +33 (0) 689 513 601 (mobile) |
|
From: Daniel P. <dp...@gm...> - 2014-06-09 05:04:11
|
Christopher, could you please get a stack trace for the logistic-regression-test failure? I can't replicate it. Dan On Mon, Jun 9, 2014 at 12:56 AM, <jen...@a2...> wrote: > Kaldi - Build # 554 - Still Failing: > > See the build log in attachment for the details. > > > > ------------------------------------------------------------------------------ > HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions > Find What Matters Most in Your Big Data with HPCC Systems > Open Source. Fast. Scalable. Simple. Ideal for Dirty Data. > Leverages Graph Analysis for Fast Processing & Easy Data Exploration > http://www.hpccsystems.com > _______________________________________________ > Kaldi-developers mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-developers > > |
|
From: <jen...@a2...> - 2014-06-09 04:54:32
|
Kaldi - Build # 554 - Still Failing: See the build log in attachment for the details. |
|
From: <jen...@a2...> - 2014-06-09 04:18:00
|
Kaldi - Build # 553 - Failure: See the build log in attachment for the details. |
|
From: <jen...@a2...> - 2014-05-28 21:12:32
|
Kaldi - Build # 540 - Fixed: See the build log in attachment for the details. |
|
From: Daniel P. <dp...@gm...> - 2014-05-28 19:34:02
|
Christopher, could you run src/ivector/logistic-regression-test from the command line and see what the error is? I can't reproduce this. Dan On Wed, May 28, 2014 at 12:05 PM, <jen...@a2...> wrote: > Kaldi - Build # 539 - Failure: > > See the build log in attachment for the details. > > > > ------------------------------------------------------------------------------ > Time is money. Stop wasting it! Get your web API in 5 minutes. > www.restlet.com/download > http://p.sf.net/sfu/restlet > _______________________________________________ > Kaldi-developers mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-developers > > |
|
From: <jen...@a2...> - 2014-05-28 19:22:51
|
Kaldi - Build # 539 - Failure: See the build log in attachment for the details. |
|
From: Daniel P. <dp...@gm...> - 2014-05-28 03:45:06
|
There is no way to do this from the command line, using current tools. You could copy and modify gmm-align-compiled to work out the scores for each frame. Dan On Wed, May 21, 2014 at 11:57 PM, Saman Mousazadeh <smo...@gm...>wrote: > To whom it may concern, > I am using Kaldi's to gmm-align-compiled to align utterances. I want > find out what is the score (likelihood) for each frame not for each > utterance. Is there any straight-forward way to do that. Right > know gmm-align-compiled gust give the likelihood for the whole utterance. > Best regards > Saman > > > > ------------------------------------------------------------------------------ > Time is money. Stop wasting it! Get your web API in 5 minutes. > www.restlet.com/download > http://p.sf.net/sfu/restlet > _______________________________________________ > Kaldi-developers mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-developers > > |
|
From: Elodie G. <elo...@gm...> - 2014-05-26 09:15:06
|
Hi, I am a student in internship and I am currently working on a project where I have to adapt Kaldi on mobile devices. I was wondering if anyone thinks that it is possible and if you got an idea how long it might take ? Does anyone tried to do that ? Thank you a lot for your answers, Elodie |
|
From: Saman M. <smo...@gm...> - 2014-05-22 06:58:06
|
To whom it may concern, I am using Kaldi's to gmm-align-compiled to align utterances. I want find out what is the score (likelihood) for each frame not for each utterance. Is there any straight-forward way to do that. Right know gmm-align-compiled gust give the likelihood for the whole utterance. Best regards Saman |
|
From: Daniel P. <dp...@gm...> - 2014-05-21 16:51:56
|
> > thanks for your reply. I've checked and finally found my mistake. I've > taken the usage "Usage: $0 <data-train> *<data-dev>* <lang-dir> > <ali-train> <ali-dev> <exp-dir>" too literally and provided the data of my > development set as "<data-dev>". > > The usage description might be a bit confusing at this point, but luckily > I figured it out. > Karel, maybe a line could be added like "e.g.: $0 data/train data/dev data/lang exp/tri3 exp/tri3_dev ... " or whatever [I'm not sure what the example would be exactly.]. I find this clarifies the usage messages. > > I've got one other question: > > I've noticed that there's only 1 CPU core used while training with a GPU. > Since it's always at 100% utilization, wouldn't it make sense to use the > other available cores as well? Is there a multithreading option? > Multi-threading isn't an easy option to use with GPUs since the "GPU context" is tied to the thread. Anyway, it shouldn't really be CPU-bound; I'd hope it was at a bit less than 100% CPU. Make sure that it's actually using the GPU. If not, it could be that you didn't compile for GPU (e.g. when you configured, nvcc was not on the command line), or that you are running with --use-gpu=no or something which turns off the GPU. It should tell you the name of the GPU it's using in the program's output, and also nvidia-smi should confirm that you are using the GPU. Dan > > 2014-05-20 17:25 GMT+02:00 Vesely Karel <ive...@fi...>: > > Hard to say what went wrong. Can you double check that the alignments >> for the corresponding utterances were correctly generated? >> (for example by ali-to-pdf with ascii output and grepping for specific >> key) >> If no targets are found, none of the frames get scored for cross-entropy. >> K. >> >> >> >> On 05/20/2014 04:47 PM, Daniel Povey wrote: >> >> I can't answer this offhand as I'm not familiar with Karel's scripts. >> Karel might be able to answer. >> Dan >> >> >> >> On Tue, May 20, 2014 at 10:45 AM, Patrick Proba <pat...@gm...>wrote: >> >>> Hello Dan, >>> >>> thanks for your answer. >>> You're right, there were only warnings in the main log file. >>> I've attached the log file where I found actual errors >>> (iter00.initial.log). >>> >>> The error which appears twice in a row is: >>> >>>> ERROR (nnet-train-frmshuff:Max():kaldi-vector.cc:515) Empty vector >>> >>> >>> The warnings leading to this seem to be: >>> >>>> WARNING (nnet-train-frmshuff:main():nnet-train-frmshuff.cc:133) *, >>>> missing targets >>> >>> >>> There are 630 IDs listed as missing targets. This is the size of my >>> dev set. Either I provided a wrong directory or I've missed a step. >>> Can you tell from the log file where the script is expecting data of >>> these utterances that's missing? >>> My guess would be that it expects to find alignments for the dev set in >>> $ali but doesn't find them, is this correct? >>> >>> Thanks in advance, >>> Patrick >>> >>> >>> 2014-05-20 16:23 GMT+02:00 Daniel Povey <dp...@gm...>: >>> >>> It looks to me like the warnings in the file you attached are not >>>> really errors. Most likely there is an error in one of the log files >>>> created by train_nnet_scheduler.sh. Look for the most recent file in >>>> $dir/log. >>>> Dan >>>> >>>> >>>> >>>> On Tue, May 20, 2014 at 8:29 AM, Patrick Proba < >>>> pat...@gm...> wrote: >>>> >>>>> Dear Kaldi developers, >>>>> >>>>> after calling the script "steps/nnet/train.sh" with the following >>>>> parameters, I get the message "run.pl: job failed, log is in >>>>> exp/dnn5b_pretrain-dbn_dnn/log/train_nnet.log" in the section "# RUNNING >>>>> THE NN-TRAINING SCHEDULER". >>>>> >>>>> "run.pl" $dir/log/train_nnet.log steps/nnet/train.sh >>>>> --feature-transform $feature_transform --dbn $dbn --hid-layers 0 >>>>> --learn-rate 0.008 data-fmllr-tri4b/train data-fmllr-tri4b/dev data/lang >>>>> $ali $ali $dir >>>>> >>>>> $dir = exp/dnn5b_pretrain-dbn_dnn >>>>> $feature_transform = exp/dnn5b_pretrain-dbn/final.feature_transform >>>>> $dbn = exp/dnn5b_pretrain-dbn/6.dbn >>>>> $ali = exp/tri4b_ali >>>>> >>>>> I can't figure out why I get this error message nor why it says that >>>>> it's manually selecting the CPU in the log file which I've attached. >>>>> >>>>> System specs: >>>>> - Ubuntu 14.04 LTS >>>>> - NVIDIA GTX 780, driver 331.62, CUDA V6.0.1 >>>>> - Intel Core i5-3570 >>>>> - 16GB RAM >>>>> >>>>> Any help would be greatly appreciated. >>>>> >>>>> Best regards, >>>>> Patrick >>>>> >>>>> >>>>> ------------------------------------------------------------------------------ >>>>> "Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE >>>>> Instantly run your Selenium tests across 300+ browser/OS combos. >>>>> Get unparalleled scalability from the best Selenium testing platform >>>>> available >>>>> Simple to use. Nothing to install. Get started now for free." >>>>> http://p.sf.net/sfu/SauceLabs >>>>> _______________________________________________ >>>>> Kaldi-developers mailing list >>>>> Kal...@li... >>>>> https://lists.sourceforge.net/lists/listinfo/kaldi-developers >>>>> >>>>> >>>> >>> >> >> > |
|
From: Jan T. <af...@ce...> - 2014-05-21 14:57:39
|
Kaldi has makefiles. The program make (and makefiles) do exactly what you need: after modifying a file, only this file and all files depending on that file will be compiled. y. On Mon, May 19, 2014 at 6:52 AM, Saman Mousazadeh <smo...@gm...>wrote: > To whom it may concern, > Is there any way to change a specific c++ code in Kaldi and then compile > it, without needing to compile all the codes in Kaldi from the begining? I > want to change a piece of code and recompile only that cc file (not all > cc file in Kaldi) > Best > Saman > > > > ------------------------------------------------------------------------------ > "Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE > Instantly run your Selenium tests across 300+ browser/OS combos. > Get unparalleled scalability from the best Selenium testing platform > available > Simple to use. Nothing to install. Get started now for free." > http://p.sf.net/sfu/SauceLabs > _______________________________________________ > Kaldi-developers mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-developers > > |
|
From: Amin J. <ami...@gm...> - 2014-05-21 10:21:13
|
23 messages has been excluded from this view by a project administrator.
