CosyVoice3 TTS
Alibaba FunAudioLLM's latest multilingual model with ~150ms bi-streaming, instruction control, and zero-shot cloning.
Wrap your text in SSML tags for precise control:
<speak><prosody rate="slow">Slow speech</prosody></speak>
Tags the selected model understands — click to drop one into your text where it happens:
This model reads plain text, so inline tags are ignored. For tag-based emotion, switch to an expressive model like Orpheus or Bark.
Define custom pronunciations (word = pronunciation):
About CosyVoice3
CosyVoice3 is the newest generation from Alibaba's FunAudioLLM team and a clear step up from CosyVoice 2. It introduces bi-streaming inference with roughly 150ms latency and instruction-based control, letting you steer emotion, speed, and volume through prompts. Speaker similarity for zero-shot voice cloning is improved, and coverage spans 9 languages plus 18 Chinese dialects. An RL-tuned variant pushes prosody to a state-of-the-art level. With a 5,000-character ceiling, fast generation, and strong cloning, it's geared toward multilingual production TTS and real-time applications.
Best for: Multilingual production TTS, real-time applications, voice cloning
Browse all CosyVoice3 voicesAt a glance
- Developer
- Alibaba (FunAudioLLM)
- License
- Apache 2.0
- Tier
- standard
- Speed
- fast
- Voice cloning
- Yes
- Languages
- English, Chinese, Japanese, Korean, German, Spanish, French, Italian, Russian
- Max characters
- 5000