GET

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

kind inputactivity audiohash 2a8b271f…extracted Nov 5, 2026version currentsource upstream spec ↗

Request body

11 properties · 1 required · 6 $defs
propertytypedescriptionconstraints
input*stringText to generate a speech from
modelstringnullable
metadataMetadataDictCustom type for metadata with embedded validation.nullable
streambooleandefault: false
prompt_cache_keystringnullable
voice_idstringThe preset or custom voice to use for generating the speech.nullable
ref_audiostring | stringThe audio reference for generating the speech.nullable
expressivenessnumberSpeaker flow-matching classifier-free guidance scale. Must be finite and at least 1.0.1 ≤ n ≤ 5 · nullable
phoneme_hintsobjectOptional pronunciation hints mapping a (case-sensitive) word to its ordered list of IPA phones.nullable
conversation_historySpeechUserTurn | SpeechAssistantTurn[]Prior turns, oldest first, alternating and ending on a user turn: `input` is the assistant reply to it. Conditions delivery, not speaker: set `voice_id` to choose the speaker. Requires `voice_id` to be a preset voice whose `kind` is `pvc`. Cannot be combined with `ref_audio`.nullable
response_formatenumOutput audio format. Defaults to mp3.pcm | wav | mp3 | flac | opus

Validate a payload

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