I run this command "online-audio-server-decode-faster" at server, and i run this command "online-audio-client --htk --vtt localhost 5010 scp:test.scp" at client.
So, i can get the result and time information.
But i want to get alignment score.
So, How can i get the alignment score by kaldi recognizer? thanks.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I run this command "online-audio-server-decode-faster" at server, and i
run this command "online-audio-client --htk --vtt localhost 5010
scp:test.scp" at client.
So, i can get the result and time information.
But i want to get alignment score.
So, How can i get the alignment score by kaldi recognizer? thanks.
Such as htk, the cammand "HVite -l * -b silence -C config -a -H hmm7/macros-i aligned.mlf -m -t 250.0 -y lab -I words.mlf -S train.scp dict monophones1" can get some information. The information contain time information, phone, alignment score and word.
An example of the results is as follows:
!MLF!
"*/F110101_0_01.lab"
0 480000 sil -218.591125 k00000000
480000 1440000 j -454.224609 k00000001
1440000 3840000 in -1066.226440
3840000 5280000 sh -701.459900
5280000 7520000 eng -1081.869141
7520000 10080000 er -1328.646362 k00000002
10080000 11200000 h -573.749634
11200000 13600000 ao -1128.040161
How do i get the similar result in online recognizers at kaldi? Thanks.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I think what you're talking about is aligning transcribed data.
The top-level script you should look at is steps/align.sh or
steps/align_si.sh, although the output it produces (e.g.
exp/tri4a_ali/ali.1.gz) is also produced by the standard training scripts
such as steps/train_lda_mllt.sh.
The program show-alignments can be used to view the alignments in a
human-readable form, and ali-to-phones has various options for converting
it to phones in a more computer-readable form; you might find the command
utils/int2sym.pl -f 2- data/lang/phones.txt < (input) > (output) useful
here.
Such as htk, the cammand "HVite -l * -b silence -C config -a -H
hmm7/macros-i aligned.mlf -m -t 250.0 -y lab -I words.mlf -S train.scp dict
monophones1" can get some information. The information contain time
information, phone, alignment score and word.
An example of the results is as follows:
!MLF!
"*/F110101_0_01.lab"
0 480000 sil -218.591125 k00000000
480000 1440000 j -454.224609 k00000001
1440000 3840000 in -1066.226440
3840000 5280000 sh -701.459900
5280000 7520000 eng -1081.869141
7520000 10080000 er -1328.646362 k00000002
10080000 11200000 h -573.749634
11200000 13600000 ao -1128.040161
How do i get the similar result in online recognizers at kaldi? Thanks.
Hi, everyone
I want to get alignment score from online recognizer.
How do i do it? Thank.
You'll have to be more specific in your question; what is the name of the
decoder you're using?
Dan
On Wed, Aug 13, 2014 at 12:00 AM, jesjkl jesjkl@users.sf.net wrote:
Hi, Daniel Povey
I run this command "online-audio-server-decode-faster" at server, and i run this command "online-audio-client --htk --vtt localhost 5010 scp:test.scp" at client.
So, i can get the result and time information.
But i want to get alignment score.
So, How can i get the alignment score by kaldi recognizer? thanks.
You would have to change the code; you can do something like
Dan
On Wed, Aug 13, 2014 at 12:16 AM, jesjkl jesjkl@users.sf.net wrote:
Hi, Daniel Povey
Thanks.
The score get from acoustic model.
But i dont have transfer the standard answer from client to server.
I want to ask how to do transmission of wav data and the answer from client to server.
And then, i can get the score of alignment from server. Thanks.
I'm sorry, I don't think I can answer that question - it's too hard to
understand and also, I think, too general.
Dan
On Mon, Sep 8, 2014 at 10:25 PM, jesjkl jesjkl@users.sf.net wrote:
Hi, Daniel Povey
Sorry, my english is very poor.
Such as htk, the cammand "HVite -l * -b silence -C config -a -H hmm7/macros-i aligned.mlf -m -t 250.0 -y lab -I words.mlf -S train.scp dict monophones1" can get some information. The information contain time information, phone, alignment score and word.
An example of the results is as follows:
!MLF!
"*/F110101_0_01.lab"
0 480000 sil -218.591125 k00000000
480000 1440000 j -454.224609 k00000001
1440000 3840000 in -1066.226440
3840000 5280000 sh -701.459900
5280000 7520000 eng -1081.869141
7520000 10080000 er -1328.646362 k00000002
10080000 11200000 h -573.749634
11200000 13600000 ao -1128.040161
How do i get the similar result in online recognizers at kaldi? Thanks.
I think what you're talking about is aligning transcribed data.
The top-level script you should look at is steps/align.sh or
steps/align_si.sh, although the output it produces (e.g.
exp/tri4a_ali/ali.1.gz) is also produced by the standard training scripts
such as steps/train_lda_mllt.sh.
The program show-alignments can be used to view the alignments in a
human-readable form, and ali-to-phones has various options for converting
it to phones in a more computer-readable form; you might find the command
utils/int2sym.pl -f 2- data/lang/phones.txt < (input) > (output) useful
here.
Dan
On Tue, Sep 9, 2014 at 1:59 AM, jesjkl jesjkl@users.sf.net wrote: