Description
Narrating a video in several languages usually means robotic delivery and a voice that changes with each language. CosyVoice is a speech generation model from Tongyi Lab that clones a voice from a few seconds of reference audio and speaks other languages in it, with full inference, training and deployment code. Streaming output suits low latency dialogue and dubbing. Apache-2.0.
Zero shot voice cloning: mimic a timbre from a short reference clip.
Multilingual and dialects: Chinese, English, Japanese, Korean and several Chinese dialects.
Emotion and instruction control: steer tone, speed and emotion with text instructions.
Streaming synthesis: low first packet latency.
Full stack code: inference, training and deployment tooling.
Features
Zero shot voice cloning: mimic a timbre from a short reference clip.
Multilingual and dialects: Chinese, English, Japanese, Korean and several Chinese dialects.
Emotion and instruction control: steer tone, speed and emotion with text instructions.
Streaming synthesis: low first packet latency.
Full stack code: inference, training and deployment tooling.
