Reputation: 21
I installed Pocketsphinx on Debian machine from official repositories. I tried to train default acoustic model with my voice (Indian english). Results are highly inaccurate.
Upvotes: 2
Views: 1089
Reputation: 25220
Pocketsphinx accuracy tuning is explained in tutorial. For Indian English you will have to build a model, model training is explained in other part of the tutorial.
You need to collect Indian English samples, the more the better. Something like 1 hour of speech is enough for single speaker. For many speakers you need more hours. Someone has to do it for Indian languages, I would not hope that someone outside will implement it for free.
Upvotes: 0
Reputation: 1706
According to Google Next conference on last Wednesday they opened their Speech to text Apis for public using with supporting of 80 language and will work with any application in real-time streaming or batch mode .check this article, i recommend it to use it return accurate results than Pocketsphinx .or check Microsoft Oxford project another very good solution
Upvotes: 0