Gemini 3.8 Flash TTS

Google’s highest-quality text-to-speech model, with inline vocal bursts and two-speaker dialogue. Released September 2026.

Model

google/gemini-3.8-flash-tts

Parameters

Parameter	Type	Required	Default	Description
`input`	string	Yes*	-	Text to speak. Language is auto-detected (130+ languages). *Provide exactly one of `input`, `text`, or `dialogue`
`text`	string	No	-	Alias for `input`
`dialogue`	array	No	-	Dialogue lines as `[{"text", "voice", "speaker"?}, ...]` — see Multi-Speaker Dialogue
`voice`	string	No	`alloy`	Voice name — a Gemini voice or an OpenAI-style alias (see Voices)
`response_format`	string	No	`mp3`	Output container: `mp3`, `wav`, `pcm`, `opus`, `aac`, `flac`
`speed`	number	No	`1.0`	Playback speed multiplier, `0.25`–`4.0`
`instructions`	string	No	-	Style directions the voice follows but does not read, e.g. `"speak like a sports commentator"`
`language`	string	No	auto	BCP-47 code to pin the output language (see Languages)
`speakers`	array	No	-	Two-speaker dialogue config (see Multi-Speaker Dialogue). When set, `voice` is ignored

Vocal Bursts

Gemini 3.8 TTS places short non-speech sounds written in angle brackets at the exact point they appear in input:

{ "input": "Did you see that? <laugh> I can't believe it worked. <long pause> Okay, again." }

Supported bursts include <laugh>, <sigh>, <cough>, <breath>, <gasp>, <short pause> and <long pause>. In two-speaker scripts, listener sounds go between pipes, like |mhm| or |yeah|. Bursts are performed, not spoken. The square-bracket tags of Gemini 3.1 Flash TTS ([whispering], [excited]) are not part of the 3.8 syntax. Use instructions for the overall delivery style instead.

Multi-Speaker Dialogue

Generate a conversation between up to 2 voices in one request. Two equivalent forms are accepted. Form 1 — dialogue array. Each line carries its own voice; lines are spoken in order:

{
  "dialogue": [
    { "text": "Did the deploy finish?", "voice": "onyx", "speaker": "Joe" },
    { "text": "Yes, all green — we are live!", "voice": "nova", "speaker": "Jane" },
    { "text": "Great news.", "voice": "onyx", "speaker": "Joe" }
  ]
}

speaker is optional — lines sharing a voice are treated as the same speaker automatically. If every line uses the same voice, the result is ordinary single-voice speech. Form 2 — script plus speakers map. Write the script in input with speaker names, and map each name to a voice:

{
  "input": "Joe: Did the deploy finish? Jane: Yes, all green — we are live!",
  "speakers": [
    { "speaker": "Joe", "voice": "onyx" },
    { "speaker": "Jane", "voice": "Leda" }
  ]
}

Speaker names must match the names used in input exactly. Both forms are limited to 2 distinct voices — a third voice is rejected with a 400.

Voices

30 prebuilt voices. Use either the Gemini name or the OpenAI-style alias in voice and speakers[].voice:

Voice	Style	OpenAI alias
`Achernar`	Soft	-
`Achird`	Friendly	-
`Algenib`	Gravelly	-
`Algieba`	Smooth	-
`Alnilam`	Firm	-
`Aoede`	Breezy	`fable`
`Autonoe`	Bright	-
`Callirrhoe`	Easy-going	`sage`
`Charon`	Informative	`onyx`
`Despina`	Smooth	-
`Enceladus`	Breathy	-
`Erinome`	Clear	-
`Fenrir`	Excitable	`ash`
`Gacrux`	Mature	-
`Iapetus`	Clear	-
`Kore`	Firm	`alloy`
`Laomedeia`	Upbeat	-
`Leda`	Youthful	`nova`
`Orus`	Firm	`verse`
`Puck`	Upbeat	`echo`
`Pulcherrima`	Forward	-
`Rasalgethi`	Informative	-
`Sadachbia`	Lively	-
`Sadaltager`	Knowledgeable	-
`Schedar`	Even	-
`Sulafat`	Warm	`coral`
`Umbriel`	Easy-going	`ballad`
`Vindemiatrix`	Gentle	-
`Zephyr`	Bright	`shimmer`
`Zubenelgenubi`	Casual	-

Languages

Language is auto-detected from input (130+ languages). Set language only to force a specific locale:

Code	Language
`ar-EG`	Arabic (Egyptian)
`bn-BD`	Bengali
`de-DE`	German
`en-IN`	English (India)
`en-US`	English (US)
`es-US`	Spanish (US)
`fr-FR`	French
`hi-IN`	Hindi
`id-ID`	Indonesian
`it-IT`	Italian
`ja-JP`	Japanese
`ko-KR`	Korean
`mr-IN`	Marathi
`nl-NL`	Dutch
`pl-PL`	Polish
`pt-BR`	Portuguese (Brazil)
`ro-RO`	Romanian
`ru-RU`	Russian
`ta-IN`	Tamil
`te-IN`	Telugu
`th-TH`	Thai
`tr-TR`	Turkish
`uk-UA`	Ukrainian
`vi-VN`	Vietnamese

Output Formats

Format	Description
`mp3`	MP3 (default)
`wav`	WAV, 24kHz 16-bit
`pcm`	Raw PCM, 24kHz 16-bit mono
`opus`	Opus in an Ogg container
`aac`	AAC (ADTS)
`flac`	FLAC lossless

Resources

List voices and languages from the Resources API:

GET /v1/resources/google/tts/voices
GET /v1/resources/google/tts/languages

Example

curl -X POST https://api.unifically.com/v1/tasks \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "google/gemini-3.8-flash-tts",
    "input": {
      "input": "Welcome back! <laugh> Your order has shipped and arrives tomorrow.",
      "voice": "nova",
      "response_format": "mp3",
      "instructions": "Speak cheerfully, like a friendly support agent."
    }
  }'

Response

{
  "code": 200,
  "success": true,
  "data": {
    "task_id": "abc123def456",
    "status": "processing"
  }
}

Completed Response

Poll GET /v1/tasks/{task_id} until status is completed:

{
  "code": 200,
  "success": true,
  "data": {
    "task_id": "abc123def456",
    "status": "completed",
    "output": {
      "audio_url": "https://files.unifically.com/audio/abc123.mp3"
    }
  }
}

Volume discounts are available for almost all models. Reach out at contact@unifically.com to discuss custom pricing.

​Model

​Parameters

​Vocal Bursts

​Multi-Speaker Dialogue

​Voices

​Languages

​Output Formats

​Resources

​Example

​Response

​Completed Response

Model

Parameters

Vocal Bursts

Multi-Speaker Dialogue

Voices

Languages

Output Formats

Resources

Example

Response

Completed Response