Open jeff1evesque opened 9 years ago
After some research, it has been determined that streaming audio frames chunk by chunk to the PocketSphinx engine is acceptable, and marginally similar to passing the entire audio file (as a whole, if possible) to the engine.
IRC #cmusphinx (08/09/14 ~ 7:35am EST):
jeffreylevesque I have a websocket server up, it sends buffered audio frames to the server. What is the best way pass the buffer to the PocketSphinx Engine. Should I concatenate the buffer into one 'wav' file, or is there another clever method?
nshm jeffreylevesque: you can pass audio chunk by chunk as soon as it is received
jeffreylevesque nshm, what if the chunk is between words, how will pocketsphinx handle that?
nshm jeffreylevesque: it will concatenate chunks and wait till the word end
jeffreylevesque nshm, if I have a file that has recording of a sentence, then have that file as buffered chunks. Which will translate faster - the complete file, or passing in chunks, or will they be marginally similar?
nshm similar
We need to determine whether buffered audio frames can be sent chunk by chunk to the PocketSphinx engine. Or, whether chunks need to be intelligently concatenated before being passed to the engine (since chunks may be between words).
After this has been determined, and the whisper submodule has been added to this repository, we can implement the former findings.