This article discusses the development of a new voice conversion model called Low-latency Low-resource Voice Conversion (LLVC). This model is designed to convert speech in real-time, with a latency of under 20ms at a bitrate of 16kHz, and can run on a consumer CPU. The model uses a generative adversarial architecture and knowledge distillation to achieve this performance. The authors claim that LLVC achieves the lowest resource usage and latency of any open-source voice conversion model. The model is available on GitHub.
Publication date: 1 Nov 2023
Project Page: https://github.com/KoeAI/LLVC
Paper: https://arxiv.org/pdf/2311.00873
