WhisperLive

A self-hosted speech-to-text server transcribes live audio and recordings with Whisper, with speaker labels and Chrome, Firefox and iOS clients.

Alternative to:

WhisperLive turns speech into text as audio arrives, using OpenAI's Whisper models. It's for developers and teams who want to run transcription on their own machine or server and connect it to live audio or recorded files.

Live audio and recordings: Transcribe microphone input, audio files, and RTSP or HLS streams, with Chrome and Firefox extensions and a native iOS client available.
Multilingual transcription and translation: Detect the spoken language with multilingual models, transcribe in the original language, or translate speech into English and other target languages.
Transcript detail: Include word timestamps and confidence scores, and use faster-whisper hotwords to improve recognition of names, acronyms and specialist terms.
Speaker labels: Identify speakers in live transcripts through the optional pyannote.audio dependency, or supply known speaker names and reference recordings through the REST API.
App integration: Connect applications through an OpenAI-compatible transcription API or a streaming client that accepts audio from your own pipeline; GPU batching supports concurrent sessions.

WhisperLive is free, open source software under the MIT license. Its Python server runs locally or on a self-hosted server, with Docker support and Linux and macOS setup options. Audio goes to the server you connect to, which can run on your own machine. Supported backends include faster-whisper, TensorRT for NVIDIA GPUs, and OpenVINO for Intel CPUs and GPUs. AMD GPU support uses ROCm with faster-whisper.