ByteDanceSeed Speech
Seed Voice Replication 2.0
Voice Replication 2.0 learns a voice profile from short audio. Its guide recommends clear single-speaker WAV samples of 14–30 seconds for better results.
Technical guide
The tone-fidelity option aims to retain the sample’s style, accent and prosody. That option supports synthesis in the sample language and is unavailable through bidirectional streaming.Seed speech synthesis and voice replication API
Practical use
Practical guidance. Record a clean sample with the speaking style you want to reproduce. Compare unfamiliar words and longer sentences, since a close match on the sample text alone is a weak check.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | seed-icl-2.0 |
| Synthesis resource | seed-icl-2.0.Seed speech synthesis and voice replication API |
Related models
- Seed TTS 2.0
ByteDance · Seed Speech
Context-aware speech synthesis model with text controls for tone, emotion and pace.
Launch date not confirmedAPI