Tag
A new streaming voice conversion model, MeanVC2, has been released, supporting cross-gender and cross-language conversion with real-time performance on CPU.
This article warns that an AI voice project with 37,000 stars on GitHub could become a new weapon for telecom fraud, introducing its features such as AI cover songs, voice conversion, and real-time voice changing, as well as its easy-to-use WebUI interface.
GRAFT is a per-word pronunciation conditioning mechanism for zero-shot text-to-speech that uses a spoken sample of a target word to control its pronunciation, achieving significant improvements in target-word phoneme error rates across multiple languages while preserving speaker similarity.