The article introduces two intelligibility prediction systems derived from a noise-robust automatic speech recognition model. These systems are designed to predict speech intelligibility in noisy environments, which is crucial for the development of speech enhancement algorithms for hearing aids. The automatic speech recognition model used is pretrained with a simulated noisy speech corpus, making these systems robust to unseen scenarios. The two systems follow different approaches, one being intrusive and leveraging hidden representations of the model, the other being non-intrusive and making predictions based on derived uncertainty.
Publication date: 3 Nov 2023
Project Page: Not Provided
Paper: https://arxiv.org/pdf/2310.19817
