Skip to content

vibium listen

Capture audio from a target and return a transcript.

vibium listen [flags]
--allow-silence Return an empty transcript instead of failing when nothing is heard
--duration int Maximum milliseconds of capture, from the call (default 30000)
--language string BCP 47 tag (default en)
--model string Provider's model id
--no-transcribe Capture without a transcript; speech onset and reply latency are still reported
-o, --output string Also save the captured audio as a WAV file
--provider string Speech-to-text provider or preset (default: from the environment)
--silence int Milliseconds of trailing silence that end the reply; 0 turns endpointing off (default 1000)
--target string session, local, or a path ending in .wav (default: the session if one is open, else local)
--timeout int Milliseconds to wait for speech to start, from the call (default 10000)
Terminal window
vibium listen
# Come here, Watson.
vibium listen --timeout 5000 --json
# {"ok":true,"result":{"text":"Come here, Watson.","speech":true,"onsetMs":640,"speechEndMs":2010,"latencyMs":640,"stoppedBy":"silence",...}}
vibium listen --target reply.wav --silence 0
# Transcribes the whole file.
VIBIUM_SHERPA_URL=ws://127.0.0.1:6006 vibium listen
# come here watson
# Transcribes through a sherpa-onnx server you started, which keeps its model loaded.
vibium listen --silence 0 --duration 5000 -o five-seconds.wav
# Captures exactly five seconds and saves them.
vibium listen --no-transcribe -o reply.wav
# Heard 2.1 s of speech after 640 ms (reply.wav)
# Saves the reply without transcribing it. Transcribe it later:
vibium listen --target reply.wav