voicelistener 1.0.1

Real-time voice recognition using Silero VAD and Whisper

voicelistener

Real-time voice recognition using Silero VAD and Whisper.

Structure

voicelistener/
├── __init__.py
├── __main__.py              # CLI entry point
├── voicelistener.py         # VoiceListener class (audio + VAD + threading)
├── requirements.txt
└── transcribers/
    ├── __init__.py
    ├── whispertranscriber.py      # WhisperTranscriber class
    └── elevenlabstranscriber.py   # ElevenLabsTranscriber class

Installation

pip install voicelistener

CLI usage

# Default (local Whisper)
python -m voicelistener

# ElevenLabs (requires ELEVENLABS_API_KEY env var)
python -m voicelistener --transcriber elevenlabs

Listens to your microphone, detects speech, and prints transcriptions to stdout. Press Ctrl+C to stop.

Flag	Default	Description
--transcriber	whisper	Speech-to-text backend (whisper or elevenlabs)

Library usage

from voicelistener import VoiceListener, WhisperTranscriber, ElevenLabsTranscriber

# Local Whisper
transcriber = WhisperTranscriber(model="base.en")

# Or ElevenLabs (set ELEVENLABS_API_KEY env var)
# transcriber = ElevenLabsTranscriber()

listener = VoiceListener(transcriber=transcriber)

for text in listener:
    print(text)

Callback style

def handle(text):
    print(f"Heard: {text}")

listener = VoiceListener(
    transcriber=WhisperTranscriber(),
    on_transcription=handle,
)
listener.start()

VoiceListener options

Parameter	Default	Description
transcriber	(required)	Object with a transcribe(audio) -> str method
silence_timeout_ms	2000	Silence duration (ms) to finalize an utterance
min_utterance_ms	250	Minimum speech length to transcribe
pre_buffer_ms	150	Audio kept before VAD triggers
vad_threshold	0.5	Silero VAD confidence threshold
energy_threshold	0.005	RMS energy below which VAD is skipped
on_transcription	None	Callback invoked with each transcription
on_speech_start	None	Callback invoked when speech is detected
on_speech_end	None	Callback invoked when speech ends (silence timeout)

Custom transcriber

Implement a class with a transcribe method:

class MyTranscriber:
    def transcribe(self, audio):
        # audio is a float32 numpy array at 16kHz
        return "transcribed text"

listener = VoiceListener(transcriber=MyTranscriber())