GET

/v1/schemas/llmgateway/audio/v1%2Faudio%2Fspeech

kind inputactivity audiohash 13ac22aa…extracted Nov 20, 2026version currentsource upstream spec ↗

Request body

9 properties · 2 required · 0 $defs
propertytypedescriptionconstraints
model*stringID of the speech generation (text-to-speech) model to use.
input*stringThe text to synthesize into speech.length ≥ 1
safety_identifierstringOpaque end-user identifier stored in log params.safety_identifier. Never forwarded upstream and never replaces LLM Gateway's provider-facing identifier.length ≤ 128
userstringLegacy end-user identifier, recorded as params.safety_identifier when safety_identifier is absent and it is at most 128 characters. Never forwarded upstream.
stream_formatenumReturn binary audio (default) or `sse` speech events: `speech.audio.delta` events with base64 audio, then a `speech.audio.done` event whose `usage` carries the request's `cost`.audio | sse
voicestringThe prebuilt voice to use. Defaults to the model's default voice when omitted.
response_formatenumThe audio format of the returned audio. OpenAI models support mp3 (default), opus, aac, flac, wav and pcm. Gemini models emit PCM, so only wav (default) and pcm are supported.wav | pcm | mp3 | opus | aac | flac
speednumberPlayback speed hint. Forwarded to OpenAI models; not applied by Gemini speech models.0.25 ≤ n ≤ 4
instructionsstringOptional style/delivery instructions prepended to the input as a natural-language directive (e.g. 'Say cheerfully').

Validate a payload

before you spend tokens
shell
$ curl -X POST https://modelschemas.com/v1/validate \
    -d '{"provider":"llmgateway","endpointId":"v1/audio/speech","payload":{…}}'
→ {"valid": false, "errors": [{"path": "#", "keyword": "required", …}]}