OpenTranscription

OpenTranscription combines all major speech-to-text engines into one endpoint, making it easy to switch providers. Audio can be sent as a batch job for pre-recorded files or streamed in real-time over WebSocket. The response always includes word-level timestamps, speaker labels, and confidence scores. Available models include those from OpenAI, Deepgram, Google Cloud, and more.

A smart router selects the model based on price, speed, or accuracy. Speaker diarization is available on supported models. The service covers over a hundred languages, with auto-detection and code-switching for mixed-language audio.

The built-in Ranker performs automated word-error-rate benchmarks against golden-set audio across eight domains. It re-runs whenever a model or its pricing changes, ensuring the leaderboard reflects current behavior.

There is a browser playground for testing files, TypeScript and Python SDKs, webhooks, and a documented HTTP API. Billing is per second of audio with no subscription, using one rate card across all providers. Users who bring their own provider keys get a hundred free minutes each month and pay their provider directly afterward.