I'm quite hazed by this problem: everything goes well with the acoustic adaptation but when it's time to test the results using pocketsphinx_batch -adcin yes -cepdir C:\Users\Luca\CMUSphinx -cepext .wav -ctl C:\Users\Luca\CMUSphinx\***.fileids -lm C:\Users\Luca\CMUSphinx\en-us.lm.bin -dict C:\Users\Luca\CMUSphinx\cmudict-en-us.dict -hmm C:\Users\Luca\CMUSphinx\en-us-adapt -hyp C:\Users\Luca\CMUSphinx\test\test.hyp -samprate 8000
the return is always this
`INFO: feat.c(715): Initializing feature stream to type: '1s_c_d_dd', ceplen=13, CMN='prior', VARNORM='yes', AGC='none'
INFO: cmn.c(143): mean[0]= 12.00, mean[1..12]= 0.0
INFO: acmod.c(164): Using subvector specification 0-12/13-25/26-38
INFO: mdef.c(518): Reading model definition: C:\Users\Luca\CMUSphinx\en-us-adapt/mdef
INFO: mdef.c(531): Found byte-order mark BMDF, assuming this is a binary mdef file
INFO: bin_mdef.c(336): Reading binary model definition: C:\Users\Luca\CMUSphinx\en-us-adapt/mdef
INFO: bin_mdef.c(516): 42 CI-phone, 137053 CD-phone, 3 emitstate/phone, 126 CI-sen, 5126 Sen, 28458 Sen-Seq
INFO: tmat.c(206): Reading HMM transition probability matrices: C:\Users\Luca\CMUSphinx\en-us-adapt/transition_matrices
No ERROR, no FATAL but also no file in the dir. It seems to me that the program crash somehow after opening tmat but I can't understand why or what to do.
The result is the same even if I try with the original tmat from sourceforge so don't seem to me a corruption problem.
Is there anything else I could provide to help you?
Last edit: Luca Montanari 2020-04-29
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I'm trying to adapt the model to our keyword to enhance accuracy, currently is always too or too little sensitive.
We spoke italian but because is only a phrase of 2 words "ciao mat", we thought that the standard CMU-Dict should work fine.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thank you Nickolay, actually it have 3, two for "cia-o" and one for "mat". Maybe is the hint to be intended per word?
The problem with Precise is that we are working on mobile, so we need a voice recognition that could work both on Android and iOS, plus we wouldn't throw away all the work done with Pocketsphinx, we really kinda like it.
Last edit: Luca Montanari 2020-04-29
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Well, if you are serious about this project you need a neural spotter otherwise its not going to work reliably, you can probably try https://github.com/hyperconnect/TC-ResNet, it supports tflite and should easily work on mobile.
If you are not serious, just select a longer keyword and train better Italian model. Still, you need linux.
Last edit: Nickolay V. Shmyrev 2020-04-30
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hy guys,
I'm quite hazed by this problem: everything goes well with the acoustic adaptation but when it's time to test the results using
pocketsphinx_batch -adcin yes -cepdir C:\Users\Luca\CMUSphinx -cepext .wav -ctl C:\Users\Luca\CMUSphinx\***.fileids -lm C:\Users\Luca\CMUSphinx\en-us.lm.bin -dict C:\Users\Luca\CMUSphinx\cmudict-en-us.dict -hmm C:\Users\Luca\CMUSphinx\en-us-adapt -hyp C:\Users\Luca\CMUSphinx\test\test.hyp -samprate 8000the return is always this
`INFO: feat.c(715): Initializing feature stream to type: '1s_c_d_dd', ceplen=13, CMN='prior', VARNORM='yes', AGC='none'
INFO: cmn.c(143): mean[0]= 12.00, mean[1..12]= 0.0
INFO: acmod.c(164): Using subvector specification 0-12/13-25/26-38
INFO: mdef.c(518): Reading model definition: C:\Users\Luca\CMUSphinx\en-us-adapt/mdef
INFO: mdef.c(531): Found byte-order mark BMDF, assuming this is a binary mdef file
INFO: bin_mdef.c(336): Reading binary model definition: C:\Users\Luca\CMUSphinx\en-us-adapt/mdef
INFO: bin_mdef.c(516): 42 CI-phone, 137053 CD-phone, 3 emitstate/phone, 126 CI-sen, 5126 Sen, 28458 Sen-Seq
INFO: tmat.c(206): Reading HMM transition probability matrices: C:\Users\Luca\CMUSphinx\en-us-adapt/transition_matrices
C:\Users\Luca\CMUSphinx\sphinxtrain\bin\Release\x64>`
No ERROR, no FATAL but also no file in the dir. It seems to me that the program crash somehow after opening tmat but I can't understand why or what to do.
The result is the same even if I try with the original tmat from sourceforge so don't seem to me a corruption problem.
Is there anything else I could provide to help you?
Last edit: Luca Montanari 2020-04-29
What are you trying to achieve overall? Do you want an English model?
I'm trying to adapt the model to our keyword to enhance accuracy, currently is always too or too little sensitive.
We spoke italian but because is only a phrase of 2 words "ciao mat", we thought that the standard CMU-Dict should work fine.
Adaptation doesn't improve the accuracy of the keyword detection. For best detection keyword should have 3-5 syllables, your one is 2.
If you still want to keep your keyword, you'd better adopt something like mycroft precise. But you will need to record much more keyword samples.
Windows is not suitable for any kind of speech work.
Thank you Nickolay, actually it have 3, two for "cia-o" and one for "mat". Maybe is the hint to be intended per word?
The problem with Precise is that we are working on mobile, so we need a voice recognition that could work both on Android and iOS, plus we wouldn't throw away all the work done with Pocketsphinx, we really kinda like it.
Last edit: Luca Montanari 2020-04-29
Well, if you are serious about this project you need a neural spotter otherwise its not going to work reliably, you can probably try https://github.com/hyperconnect/TC-ResNet, it supports tflite and should easily work on mobile.
If you are not serious, just select a longer keyword and train better Italian model. Still, you need linux.
Last edit: Nickolay V. Shmyrev 2020-04-30