Bytes input to FFmpeg in speech-recognition-from-url.py about sherpa-onnx

[tempops]

Hello,
I had to provide pipe:0 input to ffmpeg to ingest bytes on the fly and provide metadata about the bytes beforehand to make it work.

Thanks!

[csukuangfj]

I am getting raw audio bytes

Everything in the computer is represented in bytes.

Given that you did not describe any metadata about the bytes, it is not possible to tell you how to do with your bytes, since
the bytes can represent any thing.

[tempops]

These bytes are from an audio stream, I am sending the bytes to be decoded by ffmpeg and then read by Sherpa_ONNX. The audio stream is from online-websocket-client-microphone.py which is sending the microphone bytes to the websocket I had given above:

recognizer = create_recognizer(args)

byte = await websocket.recv()

ffmpeg_cmd = [
"ffmpeg",
"-i",
byte,
"-f",
"s16le",
"-acodec",
"pcm_s16le",
"-ac",
"1",
"-ar",
"16000",
"-",
]

However ffmpeg does not give any output in stdout and it is stuck at the first frame

data = process.stdout.read(frames_per_read * 2)
if not data:
break

do we need to send a specific chunk of bytes as input to ffmpeg? currently it is receiving byte length of 3200

[csukuangfj]

I suggest that you save the output of ffmpeg to a file and check the file.