Speech-to-Text (Exp)
Domain: Development & Internals
Motivation
@deepseek-ai/dsh-experimental-speech-to-text is the Service Definition that selects named speech recognizers through ctx.speechToText; consumers resolve their request before execution and providers register independently. defaultProvider is required and selects an exact registered id, with no fallback that silently picks another recognizer; the ./wave helper validates canonical 16 kHz mono PCM16 WAV input.
Concrete Example
resolve() captures the provider instance, recording, and language; transcribe() rejects a withdrawn or replaced registration, and a registration disposer closes admission and aborts accepted requests. The provider package is @deepseek-ai/dsh-experimental-speech-to-text-sensevoice.
Analogy
It is a named-registry for dictation engines where the caller must name exactly which one to use.