My question arises from the only fact that I have setup a process to quickly add new recording and re-run the acoustic model generation.
I have about 500 sentences of training data per each user, the vocabulary is only around 80 words.
I have trained it with 3 users and it works well for those three users, very good accuracy
When I want to add a new user, is it possible that I get the same accruacy for the new speaker and also ensure that the existing users accuracy doesn't get diminished if the new user voice is trained withsmaller amount of data but covering the full dictionary?
Secondly for me re-running the model generation is easier rather than the adaptation process. - as my setup is geared for it.
My question is, will the final model be different in these two scenarios?
1.if adaptation is used for a new user data and merged with model
2.new user data(minimal data) is used along with the old training data and training process is re-run using sphinxtrain.
I am using the continuous model.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
My question arises from the only fact that I have setup a process to quickly add new recording and re-run the acoustic model generation.
I have about 500 sentences of training data per each user, the vocabulary is only around 80 words.
I have trained it with 3 users and it works well for those three users, very good accuracy
When I want to add a new user, is it possible that I get the same accruacy for the new speaker and also ensure that the existing users accuracy doesn't get diminished if the new user voice is trained withsmaller amount of data but covering the full dictionary?
Secondly for me re-running the model generation is easier rather than the adaptation process. - as my setup is geared for it.
My question is, will the final model be different in these two scenarios?
1.if adaptation is used for a new user data and merged with model
2.new user data(minimal data) is used along with the old training data and training process is re-run using sphinxtrain.
I am using the continuous model.