The article presents an innovative approach to dysarthric speech reconstruction (DSR) using a system called Unit-DSR. This system uses speech units and HuBERT for domain-adaptation capacity, improving training efficiency. Compared to Neural Encoder-Decoder (NED) approaches, Unit-DSR is simpler and shows better results in terms of content restoration. It shows a 28.2% relative average word error rate reduction when compared to original dysarthric speech.
Publication date: 31 Jan 2024
Project Page: https://wyj1996.github.io/Unit-DSR-demo/index.html
Paper: https://arxiv.org/pdf/2401.14664
