Whisper API is an open-source, self-hosted speech-to-text API powered by whisper.cpp, with a Deepgram-compatible interface and 912 stars on GitHub.
Whisper API gives you cloud-style transcription without the cloud. It wraps a fine-tuned Whisper ASR model, running on the efficient whisper.cpp engine, behind a high-performance API with user-level access support. Its standout design choice is compatibility: the interface mirrors Deepgram’s, including the /v1/listen endpoint style over both REST and WebSocket, so applications built against a commercial transcription API can be pointed at your own server with minimal changes — while your audio and transcripts stay under your full control. Features go beyond basic transcription, with speaker diarization, custom vocabulary through prompting, audio cropping, and exports to JSON, SRT and VTT for subtitles and captioning workflows. The project is packaged for straightforward deployment, including a Docker setup, and its documentation walks through configuration, model choices and API usage in detail. User-level access support means a single deployment can safely serve several people or applications with separate API keys.
Whisper API is free and open source under the MIT license. You self-host it — including as a Docker app — and pay only for your own compute instead of per-minute transcription fees.
Whisper API is for developers who need speech-to-text in their products but want data ownership and predictable costs. It is especially attractive if you already use a Deepgram-style API and want a self-hosted drop-in alternative, or if you process large volumes of audio where per-minute pricing adds up.