TVTSyn: Content-Synchronous Time-Varying Timbre for Streaming Voice Conversion and Anonymization

Researchers have unveiled TVTSyn, a new AI model for real-time voice conversion and anonymization. It solves a core problem by making speaker identity change dynamically with speech content, not using a static embedding. The system achieves under 80 milliseconds of latency on a GPU, outperforming current state-of-the-art methods in naturalness and speaker transfer while better preserving privacy by reducing speaker leakage in the converted audio.