Revolutionary TTS Model Allows Seamless Audio Word Replacement!

Chinese startup Yunshang Qulv has unveiled ViiTorVoice-NAR, an open-source text-to-speech model that enables precise word-level editing in audio recordings without regenerating surrounding content. Available on GitHub and Hugging Face, it boasts competitive word error rates and fast processing. Additionally, it supports reference-text-free voice cloning, raising crucial legal and ethical considerations in AI audio manipulation.