I want to create a automatic speech recognition system that will identify a correct word from a list of words in the database. I have seen CMUSphinx can be used for this problem. I have tried the hello world sphinx demo app, but it gives not expected results.
I don't know how to choose the correct acoustic model, dictionary file, language model. For a single word is the language model necessary?
Is there any prebuilt acoustic model for Indian English?
I have tried the hello world sphinx demo app, but it gives not expected results.
You need to provide more details on what have your tried. Keyword spotting is supported by pocketsphinx, you can check it here
http://cmusphinx.sourceforge.net/wiki/tutorialpocketsphinx
To test keyword spotting you can try:
pocketsphinx_continuous -infile file.wav -keyphrase "hello world" -kws_threshold 1e-40
that will look for a keyphrase in a file.
I don't know how to choose the correct acoustic model, dictionary file, language model. For a single word is the language model necessary?
Language model is not necessary. For acoustic model and the dictionary you can select the ones that best suited for your language.
Is there any prebuilt acoustic model for Indian English?
There are no models for Indian English yet but you can try US-English model, it should work ok for a single word.
来源：https://stackoverflow.com/questions/25637872/speech-to-text-for-single-word