Fish Speech
Fish Speech is an open-source text-to-speech (TTS) system based on deep neural networks, supporting multilingual speech synthesis and voice cloning. The system can extract speaker characteristics from a small number of audio samples to generate synthesized speech with natural prosody. It is applicable to voice assistants, content creation, dubbing, and other use cases, and is characterized by lightweight inference and high configurability for developers and researchers.
Timeline (last 90 days)
Mainstream STT solutions include the Whisper series, FunASR and others; in recent years TTS solutions based on diffusion models and neural codecs (such as CosyVoice and Fish Speech) have achieved naturalness close to that of real humans.
Modern open-source TTS models such as CosyVoice and Fish Speech can already achieve fine-grained control over timbre, emotion, speech rate, and pauses.
All Facts (4)
Fish Speech是一个开源的文本转语音模型，其S2 Pro版本支持零样本和少样本声音克隆，只需几秒到几十秒的参考音频即可生成高度相似的合成语音
80%UnverifiedFish Speech底层采用了VITS架构的改进版本，将文本编码、韵律预测和声码器整合在一个端到端的框架中
70%UnverifiedMainstream STT solutions include the Whisper series, FunASR and others; in recent years TTS solutions based on diffusion models and neural codecs (such as CosyVoice and Fish Speech) have achieved naturalness close to that of real humans.
50%UnverifiedModern open-source TTS models such as CosyVoice and Fish Speech can already achieve fine-grained control over timbre, emotion, speech rate, and pauses.
50%