I want FFmpeg to quit recording of mic when it detects silence below a threshold

I am trying to make a real time speech to text transcription script. The below script works for recorded wav file.

 from asrecognition import ASREngine
 asr = ASREngine("tr", model_path="mpoyraz/wav2vec2-xls-r-300m-cv6-turkish")
 audio_paths = ["prerecorded.wav"]
 transcriptions = asr.transcribe(audio_paths)
 print(transcriptions)

But I want to capture voice from microphone, when a silence level for a while is detected,
recording of voice must stop and it will be piped to speech recognition engine for transcription. Then recording of voice from mic must restart again.

I thought FFmpeg could achieve it, but how?

You could achieve this in a combination of pulseaudio and ffmpeg:

**Code based on the python pulsectl lib **

pulse = pulsectl.Pulse("Test1")

1. Retrieve the pulse.sink_input_list -which only exists if sink is present (e.g. a mic) ->pulseSinkInputInfoList.

   pulseSinkInputInfoList = pulse.sink_input_list()
   monName =pulse.sink_info(pulseSinkInputInfoList[0].sink).monitor_source_name
   sources = pulse.source_list()
   for pulseSourceInfo in sources:
    if pulseSourceInfo.name==monName:
        while True:
            mos=pulseSourceInfo.index
            peak= pulse.get_peak_sample(mos, 0.2)
            if peak > 0:
                execute ffmpeg like:
                  fmpeg -f pulse -i alsa_input.pci-0000_00_1b.0.analog-stereo -ac 1 recording.m4a
   

This is rather a stack overflow theme, therefore I kept it short - just to give you a gist of it. The base idea is, that you use pulse audio to detect some silence (or the opposite) in the "peak" line. Then you could execute an ffmpeg command to record a fragment. To my knowlege you'll end up with a bunch of short clip which you could join/concat afterwards using ffmepg's concat protocol.

Infos about ffmpegs pulse protocol and concat protocol

An implementation of that protocol in python can be found here