TOOLORA / Local reading edition
CosyVoice
Synthesize multilingual speech from text and reference audio, with streaming input/output, pronunciation correction and language, emotion or speed instructions.
Overview
CosyVoice is a deployable speech-synthesis family. Current Fun-CosyVoice 3.0 covers nine languages, reference-audio conditioning and cross-language voice reproduction. Instructions control dialect, emotion, speed and volume; Chinese pinyin or English phonemes can correct pronunciation. Streaming text input and audio output support integration. Official base and RL checkpoints, training and inference code are provided. The current model card and main code declare Apache 2.0; runtime dependencies vary by version.
Official sources and references
- Official websitehttps://github.com/QwenAudio/CosyVoice
- Documentationhttps://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
Descriptions summarize public information; consult official websites for features, pricing and regional availability.