In my experiment, I would like to detect a simple keyword, say 'glass'. I have trained the whole word HMM model for the samples of 'glass' I have collected. But how to train the garbage model? Shall I train it with noise sample or shall I train it with random words other than 'glass'?
If I have both models, during the testing I can compare the log likelihood score and make a decision.
Thanks.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Can any one give a simple example that includes phone loop, logp, and garbage words that can increase the accuracy of recognition....
Thank You in advance....
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
How to train the garbage model?
In my experiment, I would like to detect a simple keyword, say 'glass'. I have trained the whole word HMM model for the samples of 'glass' I have collected. But how to train the garbage model? Shall I train it with noise sample or shall I train it with random words other than 'glass'?
If I have both models, during the testing I can compare the log likelihood score and make a decision.
Thanks.
You need to train it both with noises and random words.
how many random words do I need?
30 mins of speech
And, the easiest way would be to use context-independent phone loop instead of garbage, you can even embed it into FSG grammar.
Can any one give a simple example that includes phone loop, logp, and garbage words that can increase the accuracy of recognition....
Thank You in advance....
Is there any post/document which explains how to implement phone loops / garbage model for OOV words