Ali Tongyi announced the open source Fun-CosyVoice 3-0.5b. This version provides zero-shot sound cloning capabilities. You only need to provide a reference audio of at least 3 seconds to reproduce the sound and synthesize the new voice, and supports local deployment and secondary development. In addition, Tongyi launched a lightweight version of the Fun-ASR-Nano model. The total number of parameters was reduced to 0.8B, and the inference cost was lower. It is now open source and supports local deployment and customized fine-tuning.

Ali Tongyi announced the open source Fun-CosyVoice 3-0.5b. This version provides zero-shot sound cloning capabilities. You only need to provide a reference audio of at least 3 seconds to reproduce the sound and synthesize the new voice, and supports local deployment and secondary development. In addition, Tongyi launched a lightweight version of the Fun-ASR-Nano model. The total number of parameters was reduced to 0.8B, and the inference cost was lower. It is now open source and supports local deployment and customized fine-tuning.