Skip to main content
Every speech-to-text model has its own route. The reference lists one page per model and protocol, generated from the gateway’s OpenAPI and AsyncAPI specs.
  • SLNG hosted: models running on SLNG infrastructure. Routes carry the slng/ prefix, for example /v1/stt/slng/deepgram/nova:3-en, and can target a specific region, as described in the regions map.
  • Third party: models called through SLNG on the provider’s infrastructure, one group per provider. Routes use the provider name, for example /v1/stt/deepgram/nova:3. Use the route directly, or register your own provider key with bring your own key.
  • Batch: asynchronous transcription jobs for recorded audio, with their own job and file endpoints.
Each model lists the protocols it supports: HTTP for a single request and transcript, WebSocket for streaming with partial results. WebSockets vs HTTP covers the choice. Pricing, languages, and regions per model are in the model catalog.