whisper-live-client

A lightweight Python client for WhisperLive,
designed for client-only use cases with significantly fewer dependencies than the official server library.

Features

• Sends PCM audio to a WhisperLive WebSocket endpoint
• Real-time transcription support with simple callback interface
• Fewer dependencies – ideal for minimal environments or custom frontends
• Focused on client-side use

Installation

pip install whisper-live-client

or install from source:

git clone https://github.com/B4dT0bi/whisper-live-client
cd whisper-live-client
pip install .

Usage

from whisper_live.transcription_client import TranscriptionClient

def my_segment_callback(segments):
    print(f"Received segments:{segments}")

    transcription_client = TranscriptionClient(
        host="localhost",         # or your server address
        port=9090,                # or your server port
        lang="en",                # or your desired language
        model="small",            # or your desired model
        use_vad=True,             # use voice activity detection
        pyaudio_input_device_id=1 # optional: specify input device ID for PyAudio
    )
    transcription_client.register_callback("process_segments", my_segment_callback) # register callback
    transcription_client() # start the client

Dependencies

This library uses only a minimal set of runtime dependencies:

• websockets – WebSocket client
• numpy – for handling audio data

Optional (for examples):

• pyaudio – audio capture in demos

This is in contrast to the official WhisperLive server code, which depends on heavy libs like torch.

📜 License

MIT License

---

Disclaimer: This is an independent client library and is not affiliated with Collabora or OpenAI.