I'm trying to train a new accoustic model for speechrecognition, with custom features. Is it possible to feed the feature matrix instead of wav file into CMUSphinx for training? I have tried searching on google and the tutorial page and can't find any information about it. Am i missing something?
Is it possible to feed the feature matrix instead of wav file into CMUSphinx for training?
It is better to train with more modern toolkit like Kaldi.
I have tried searching on google and the tutorial page and can't find any information about it. Am i missing something?
You just prepare mfc files yourself and skip first feature_extraction step. You have to write code for that like described here:
https://cmusphinx.github.io/wiki/mfcformat/
In kaldi you just prepare ark files yourself: