search

usefulsensors / openai-whisper

Robust Speech Recognition via Large-Scale Weak Supervision
MIT License
62 stars 24 forks source link

Decoding direct from microphone #23

Open abouquet opened 1 year ago

abouquet commented 1 year ago

Hi,

This POC is very interesting because we can use a language agnostic model, and use a pivot language (english) to deal some processing on it. Have you an idea if we can extends this POC to use input direct from microphone ?

I have already made a wrapper by using android.speech.SpeechRecognizer and extending SpeechService to keep sticking on Android Speech abstraction layer to use Kaldi as an implementation. It would be nice to test this 2 implementations together :)

Thanks !

nyadla-sys commented 1 year ago

@abouquet Please use below implementation to directly take the mic input and feed it to stream application that can transcribe text directly https://github.com/usefulsensors/openai-whisper/tree/main/stream_standalone

abouquet commented 1 year ago

So, you mean, that I have to implement direct streaming, in the native side and make the java wrapper around it ?