Hi all,
I generated new model for 5 words using sphinxTrain, and it works excellent in
pocketsphinx windows, but when i used this model in pocketsphinx android it is
not working (sometimes it gives poor accuracy or no recognition at all). I
have uploaded the model files in this link http://www.4shared.com/folder/Ml_C
ylL0/_online.html . it
will be very helpful if someone helps.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Android records audio at 8khz so you need to modify your training feature
extraction parameters to match this. upperf should be about 3500 and number of
filters needs to be updated accordingly.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
As of the suggestion u gave i made the models and it works. Thank u very much
Mr.Nickolay.. But still there is a strange thing happening in recognition (It
doesnt recognizes the word BACKWARD from my dictionary ). I have 5 words in
dictionary and model was generated with audio files sampled at 8khz and the
arguments are -nfilt 31,-lowerf 200,-upperf 3500 -srate 8000 and -nfft 256.
for your information i have generated the voice samples from Festvox voice
synthesis online
demo.http://festvox.org/voicedemos.html
Is there any more changes we need to do in Training arguments ?
whether the voice samples that i have generated from Festvox would make any
difference in recognition accuracy?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
whether the voice samples that i have generated from Festvox would make any
difference in recognition accuracy?
To train the model with good accuracy you need a diverse set of speech samples
to train the database. You can learn more details about training in tutorial
Hi all,
I generated new model for 5 words using sphinxTrain, and it works excellent in
pocketsphinx windows, but when i used this model in pocketsphinx android it is
not working (sometimes it gives poor accuracy or no recognition at all). I
have uploaded the model files in this link http://www.4shared.com/folder/Ml_C
ylL0/_online.html . it
will be very helpful if someone helps.
Hello
Android records audio at 8khz so you need to modify your training feature
extraction parameters to match this. upperf should be about 3500 and number of
filters needs to be updated accordingly.
thank u Mr.nickolay. for your help.... i am trying to do it..
As of the suggestion u gave i made the models and it works. Thank u very much
Mr.Nickolay.. But still there is a strange thing happening in recognition (It
doesnt recognizes the word BACKWARD from my dictionary ). I have 5 words in
dictionary and model was generated with audio files sampled at 8khz and the
arguments are -nfilt 31,-lowerf 200,-upperf 3500 -srate 8000 and -nfft 256.
for your information i have generated the voice samples from Festvox voice
synthesis online
demo.http://festvox.org/voicedemos.html
Is there any more changes we need to do in Training arguments ?
whether the voice samples that i have generated from Festvox would make any
difference in recognition accuracy?
To train the model with good accuracy you need a diverse set of speech samples
to train the database. You can learn more details about training in tutorial
http://cmusphinx.sourceforge.net/wiki/tutorial
Thank u Mr.Nickolay. I have done It and got the results.... Got a good
recognition accuracy. Thank u.