Closed ahmedalbakiri closed 7 years ago
You can find models for other supported languages in our downloads area.
So, I have already build my acoustic model and LM and dic, and the sound is in .wav. can this work there?
2016-10-05 10:58 GMT+03:00 Nickolay V. Shmyrev notifications@github.com:
Closed #34 https://github.com/cmusphinx/node-pocketsphinx/issues/34.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/cmusphinx/node-pocketsphinx/issues/34#event-812880727, or mute the thread https://github.com/notifications/unsubscribe-auth/AVlABHv4zv1sa9ZUoqvMrvTIEESJt_18ks5qw1i8gaJpZM4KOM4b .
Yes
Thanks, for your fast response,
_exports_Hypothesis { prob: -4802, bestScore: -5894, hypstr: '' }
INFO: ngram_search.c(1027): bestpath 0.00 CPU 0.000 xRT
INFO: ngram_search.c(1030): bestpath 0.00 wall 0.000 xRT
9 15
Please provide more complete data about your problem - what do you run exactly, what is exactly in the log output, what is your data file, what are your models, etc.
All you need it is in attachment.
2016-10-05 16:41 GMT+03:00 Nickolay V. Shmyrev notifications@github.com:
Please provide more complete data about your problem - what do you run exactly, what is exactly in the log output, what is your data file, what are your models, etc.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/cmusphinx/node-pocketsphinx/issues/34#issuecomment-251677732, or mute the thread https://github.com/notifications/unsubscribe-auth/AVlABDfCen4GGEEDnPJoWIrqedccd0U0ks5qw6kJgaJpZM4KOM4b .
I'm sorry, I do not see any attachments. Maybe you can simply share the file with dropbox/google drive and give here a link.
The below is the link in google drive.
https://drive.google.com/open?id=0BxN0qMy1FI89cS03OWIyemFXUHM
2016-10-05 17:38 GMT+03:00 Nickolay V. Shmyrev notifications@github.com:
I'm sorry, I do not see any attachments. Maybe you can simply share the file with dropbox/google drive and give here a link.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/cmusphinx/node-pocketsphinx/issues/34#issuecomment-251693986, or mute the thread https://github.com/notifications/unsubscribe-auth/AVlABBNpCe7VjJpnZLvM2GRJZaayniomks5qw7ZcgaJpZM4KOM4b .
Mr Nickolay, did you found where is the problem, I try to solve it, but I still get the same output
9 15
I checked your files. I do not think your model is properly trained. For example, your dictionary doesn't follow tutorial recommendation to use phones or word-specific phones. Your phones are whole syllables which is not a good idea. I also not not know how much training data do you have, what is the error rate during training and so on.
I recommend you to fix model training first and share the model training folder if you still have problems.
I train using context independent (ci), the WER and the results I have in this link,
https://drive.google.com/open?id=0BxN0qMy1FI89UHVmVHd2RlBhZTA
2016-10-09 19:21 GMT+03:00 Nickolay V. Shmyrev notifications@github.com:
I checked your files. I do not think your model is properly trained. For example, your dictionary doesn't follow tutorial recommendation to use phones or word-specific phones. Your phones are whole syllables which is not a good idea. I also not not know how much training data do you have, what is the error rate during training and so on.
I recommend you to fix model training first and share the model training folder if you still have problems.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/cmusphinx/node-pocketsphinx/issues/34#issuecomment-252495991, or mute the thread https://github.com/notifications/unsubscribe-auth/AVlABCE2mXOLguRbm-Sezp2xQb1f7jI1ks5qyRRtgaJpZM4KOM4b .
You need to use proper phoneset first. Then you need to share a whole model training folder.
The whole model is in link, and the results I have is good with these .dic file as you see in results file.
https://drive.google.com/open?id=0BxN0qMy1FI89UHVmVHd2RlBhZTA
2016-10-09 20:30 GMT+03:00 Nickolay V. Shmyrev notifications@github.com:
You need to use proper phoneset first. Then you need to share a whole model training folder.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/cmusphinx/node-pocketsphinx/issues/34#issuecomment-252499961, or mute the thread https://github.com/notifications/unsubscribe-auth/AVlABJHBpzngEWKF1Mqep19MAjQzIKPwks5qySTKgaJpZM4KOM4b .
You do not have required amount of data for training too. As I said, follow the tutorial recommendations in order to get a working system.
do you mean increase the number of sound, to train different speakers?
2016-10-09 20:54 GMT+03:00 Nickolay V. Shmyrev notifications@github.com:
You do not have required amount of data for training too. As I said, follow the tutorial recommendations in order to get a working system.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/cmusphinx/node-pocketsphinx/issues/34#issuecomment-252501310, or mute the thread https://github.com/notifications/unsubscribe-auth/AVlABAGy_Xz_nE5oddrByVBZBPSPXQScks5qySo_gaJpZM4KOM4b .
http://cmusphinx.sourceforge.net/wiki/tutorialam
- 1 hour of recording for command and control for single speaker
- 5 hour of recordings of 200 speakers for command and control for many speakers
- 10 hours of recordings for single speaker dictation
- 50 hours of recordings of 200 speakers for many speakers dictation
Thanks, I now modify the dict and got a results, but its not good results, it is recognition is bad
_exports_Hypothesis { prob: 0, bestScore: -22760, hypstr: 'لَمْ' }
INFO: ngram_search.c(1027): bestpath 0.00 CPU 0.000 xRT
INFO: ngram_search.c(1030): bestpath 0.00 wall 0.000 xRT
9 13
لَمْ 14 590
لَمْ
why its in previous previous dic produce a good results in the results folder based WER, but in the pocketsphinx node give a SIL, and now after modification to dic file give results, its not that good, but there is some output?
2016-10-09 21:14 GMT+03:00 Nickolay V. Shmyrev notifications@github.com:
http://cmusphinx.sourceforge.net/wiki/tutorialam
- 1 hour of recording for command and control for single speaker
- 5 hour of recordings of 200 speakers for command and control for many speakers
- 10 hours of recordings for single speaker dictation
- 50 hours of recordings of 200 speakers for many speakers dictation
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/cmusphinx/node-pocketsphinx/issues/34#issuecomment-252502547, or mute the thread https://github.com/notifications/unsubscribe-auth/AVlABAjdOJerML-c1g8iiANgXUv6DA39ks5qyS77gaJpZM4KOM4b .
why its in previous previous dic produce a good results in the results folder based WER, but in the pocketsphinx node give a SIL, and now after modification to dic file give results, its not that good, but there is some output?
When you do not have enough data results are unstable.
Can I used it to recognition any language or just for English ?