Seed Audio 1.0: FAQ
Seed Audio 1.0 converts text into natural-sounding speech, with multilingual preset voices, a prompt-driven generative mode, and fine control over speed, volume, pitch, and output format.
1. What is Seed Audio 1.0?
It is a text-to-speech model (ByteDance Seed Audio 1.0). Pick a preset voice to read your exact text aloud, or use Auto mode to generate audio from a scene description. It has strong multilingual and Chinese-language coverage.
2. How much text can it take at once?
Up to 5,000 characters per generation.
3. What voices are available?
Auto (prompt-driven) is the default. For verbatim reading you pick a preset: Vivi, Mindy, Kian, Cedric, Sophie, Jean, Magnus, Mabel, Nadia, Opal, Pearl, Quentin, Corinne, Esther, Lyla, Tracy, Sandy, Felix, Celeste, and Monkey King. Vivi and Mindy are the most multilingual; Felix, Celeste, and Monkey King are Chinese only. Preset voices span English, Chinese, Japanese, Spanish, Indonesian, and Portuguese, so match the voice to the text's language.
4. What is Auto mode?
With no preset voice, Seed Audio runs in generative mode. Your text is treated as a scene brief (for example a suspense radio drama) rather than lines to read verbatim, and it is enhanced into a performed scene anchored to the prompt's own language. Request a different language directly in the prompt to hear it.
5. How do I control the delivery?
Speed runs from 0.5 to 2.0 (1.0 is natural), volume from 0.5 to 2.0 (1.0 is default), and pitch shifts by semitone from -12 to +12 (0 is natural).
6. What output formats and sample rates can I get?
Formats are mp3, wav, pcm, and ogg_opus. Sample rates are 8, 16, 24, 32, 44.1, and 48 kHz; the 24 kHz default suits speech, higher rates suit music-grade output, and 8 or 16 kHz suit telephony.
7. What is it not for?
It has no expressive audio tags (for tag-driven delivery use ElevenLabs v3), reference-audio voice cloning is not exposed, and it is not for singing (use the music model).
At a glance
Axis | Detail |
|---|---|
Purpose | Text-to-speech, plus prompt-driven generative audio |
Text length | Up to 5,000 characters |
Voices | Auto (default) plus 20 presets across 6 languages |
Auto mode | Prompt becomes a performed scene, not verbatim text |
Speed | 0.5 to 2.0 (default 1.0) |
Volume | 0.5 to 2.0 (default 1.0) |
Pitch | -12 to +12 semitones (default 0) |
Formats | mp3, wav, pcm, ogg_opus |
Sample rates | 8 to 48 kHz (default 24 kHz) |
Not for | Audio tags, voice cloning, singing |
Seed Audio 1.0 vs ElevenLabs v3
Both turn text into speech. Seed Audio adds a prompt-driven Auto scene mode, pitch and sample-rate control, and stronger Chinese and multilingual coverage. ElevenLabs v3 adds expressive audio tags like [excited] and [whisper] and stability, similarity, and style controls. Choose Seed Audio for multilingual or generative-scene work, and ElevenLabs v3 for tag-driven expressive narration.