面壁智能发布VoxCPM2,一个2B参数的开源语音模型,支持30种语言及9种方言。该模型实现“声音可编辑”:通过提示词指定年龄、音色、情绪和语速,也可上传参考音频保留原音色并重新控制表达方式。实测显示,语音生成已从单纯模仿真人转向按需导演级表演,让声音变得像图片滤镜一样可描述、复制和改写。
Big thanks for this fantastic share and hands-on testing of VoxCPM2! 👍 Voice is becoming editable - that's the shift we're driving. With VoxCPM2, you get Voice Design + Controllable Cloning, 30 languages &; 9 dialects, all in a 2B open-source model. https://github.com/OpenBMB/VoxCPM