Text-to-speech

Trussium exposes provider-neutral, non-streaming text-to-speech at POST /v1/audio/speech. Applications register a SpeechCapability implementation in the sealed capability registry.

Requests contain bounded text, model, voice, output format, and speed options. Responses contain a base64-encoded audio artifact and provider metadata. Input text, audio bytes, credentials, endpoints, and provider payloads are not logged.

Self-hosted Piper

For self-hosted text-to-speech, install the independently versioned trussium-provider-piper package and register PiperSpeechCapability in the application. Configure its base_url to the separately operated Piper HTTP service, for example http://piper:5000. The adapter sends text, voice, and speed to Piper and normalizes the returned audio to the runtime's base64-encoded WAV response. It does not install, configure, or manage Piper, and non-WAV output is rejected.