Gemini TTS

POST/v1/run

Gemini TTS by Google - convert text to natural-sounding speech with AI. Supports multiple voices, languages, emotions, and speaking styles for content creation and accessibility.

Request body

Submit an async generation request. The model field selects the model; other fields are model-specific input parameters.

stringmodelrequired

Model identifier. Set to google/gemini-tts.

Default: google/gemini-tts

stringpromptrequired

The text to convert to speech. Gemini TTS supports natural-language prompting for style, pace, accent, and emotional expression — include delivery instructions inline with the text (e.g. 'Say cheerfully: Have a wonderful day!'). For multi-speaker synthesis, prefix lines with speaker aliases defined in the speakers field (e.g. 'Alice: Hello!\nBob: Hi!'). Supports inline pace/style markers like [slowly], [whispering], [excited], [extremely fast].

Optional<string>output_format

Audio output format. mp3: compressed, small file size (recommended). wav: uncompressed PCM wrapped in WAV (24 kHz, 16-bit mono). ogg_opus: Ogg container with Opus codec, good quality-to-size ratio.

Allowed values: wav, mp3, ogg_opus

Default: mp3

Optional<string>style_instructions

Optional style and delivery instructions prepended to the prompt. Controls expressiveness, accent, pace, tone, and emotional expression using natural language. Use this to separate style control from the text content. Examples: 'Speak warmly and slowly', 'Read this as a dramatic newscast', 'Use a British accent with a cheerful tone', 'Whisper mysteriously'.

Optional<string>language_code

Language for multilingual synthesis. When set, steers the model to speak in the specified language. Supports 24 GA languages and 60+ Preview languages. If not set, the model auto-detects the language from the text.

Allowed values: Arabic (Egypt), Bangla (Bangladesh), Dutch (Netherlands), English (India), English (US), French (France), German (Germany), Hindi (India), Indonesian (Indonesia), Italian (Italy), Japanese (Japan), Korean (South Korea), Marathi (India), Polish (Poland), Portuguese (Brazil), Romanian (Romania), Russian (Russia), Spanish (Spain), Tamil (India), Telugu (India), Thai (Thailand), Turkish (Turkey), Ukrainian (Ukraine), Vietnamese (Vietnam), Afrikaans (South Africa), Albanian (Albania), Amharic (Ethiopia), Arabic (World), Armenian (Armenia), Azerbaijani (Azerbaijan), Basque (Spain), Belarusian (Belarus), Bulgarian (Bulgaria), Burmese (Myanmar), Catalan (Spain), Cebuano (Philippines), Chinese Mandarin (China), Chinese Mandarin (Taiwan), Croatian (Croatia), Czech (Czech Republic), Danish (Denmark), English (Australia), English (UK), Estonian (Estonia), Filipino (Philippines), Finnish (Finland), French (Canada), Galician (Spain), Georgian (Georgia), Greek (Greece), Gujarati (India), Haitian Creole (Haiti), Hebrew (Israel), Hungarian (Hungary), Icelandic (Iceland), Javanese (Java), Kannada (India), Konkani (India), Lao (Laos), Latin (Vatican City), Latvian (Latvia), Lithuanian (Lithuania), Luxembourgish (Luxembourg), Macedonian (North Macedonia), Maithili (India), Malagasy (Madagascar), Malay (Malaysia), Malayalam (India), Mongolian (Mongolia), Nepali (Nepal), Norwegian Bokmal (Norway), Norwegian Nynorsk (Norway), Odia (India), Pashto (Afghanistan), Persian (Iran), Portuguese (Portugal), Punjabi (India), Serbian (Serbia), Sindhi (India), Sinhala (Sri Lanka), Slovak (Slovakia), Slovenian (Slovenia), Spanish (Latin America), Spanish (Mexico), Swahili (Kenya), Swedish (Sweden), Urdu (Pakistan)

Optional<array<object>>speakers

Multi-speaker voice configuration. When set, enables multi-speaker synthesis where different parts of the text are spoken by different voices. Each speaker needs a voice and a speaker_id (alias) that matches prefixes in the prompt. Requires gemini-2.5-pro-tts or gemini-2.5-flash-tts model. Not supported with gemini-2.5-flash-lite-preview-tts.

Optional<string>model

Which Gemini TTS model to use. gemini-2.5-flash-tts: low latency, cost-efficient for everyday applications (recommended). gemini-2.5-pro-tts: highest quality, best for structured workflows like podcasts, audiobooks, and customer support.

Allowed values: gemini-2.5-flash-tts, gemini-2.5-pro-tts

Default: gemini-2.5-flash-tts

Optional<string>voice

Voice preset for single-speaker synthesis. 30 distinct voices are available. Ignored when speakers is set. Popular choices: Kore (strong, firm female), Puck (upbeat, lively male), Charon (calm, professional male), Zephyr (bright, clear female), Aoede (warm, melodic female).

Allowed values: Achernar, Achird, Algenib, Algieba, Alnilam, Aoede, Autonoe, Callirrhoe, Charon, Despina, Enceladus, Erinome, Fenrir, Gacrux, Iapetus, Kore, Laomedeia, Leda, Orus, Pulcherrima, Puck, Rasalgethi, Sadachbia, Sadaltager, Schedar, Sulafat, Umbriel, Vindemiatrix, Zephyr, Zubenelgenubi

Default: Kore

Optional<number>temperature

Controls the randomness of the speech output. Higher values produce more creative and varied delivery, while lower values make the output more predictable and focused.

Range: 0 to 2

Default: 1

Response Schema

The submit endpoint returns an accepted generation task. Poll the result endpoint with the returned id for terminal outputs or errors.

Optional<string>error

Error message if the task failed. Empty on success.

stringidrequired

Unique identifier for the generation task.

Optional<string>model

Model ID used for the prediction.

Optional<array>outputs

Array of generated content. Empty when status is not completed.

stringstatusrequired

Status of the task: pending, running, completed, failed, or timeout.

Allowed values: pending, running, completed, failed, timeout

Model capabilities

array<string>capability_tagsrequired

Capabilities declared by the model registry.

Default: text-to-speech

stringexecution_moderequired

Execution mode declared by the model registry.

Default: async

# 1. Submit a generation request
curl -X POST https://api.sandbase.ai/v1/run \
  -H "Authorization: Bearer $SANDBASE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "google/gemini-tts",
  "prompt": "Host: Welcome back to AI Frontiers, the podcast where we explore the latest breakthroughs in artificial intelligence. Today we have a very special guest. Doctor Chen, thank you for joining us!\nDrChen: Thanks for having me! I'm excited to be here.\nHost: So, let's dive right in. Your recent paper on neural architecture search has been making waves. Can you tell our listeners what inspired this research?\nDrChen: Absolutely. It all started when we noticed that most existing approaches were optimizing for the wrong metrics. We asked ourselves, what if we could let the model design itself?",
  "output_format": "mp3",
  "voice": "Kore"
}'

# 2. If the response is still running, poll the returned id
curl https://api.sandbase.ai/v1/run/run_abc123 \
  -H "Authorization: Bearer $SANDBASE_API_KEY"

Response202 Accepted

{
  "id": "run_abc123",
  "model": "google/gemini-tts",
  "status": "running",
  "outputs": []
}