Toward Universal Speech Enhancement for Diverse Input Conditions

This article presents research into the development of a universal speech enhancement model that can handle diverse input conditions. Existing techniques have shown strong performance but are typically designed for a single condition or task. The proposed model is independent of microphone channels, signal lengths, and sampling frequencies, and it combines existing public corpora with multiple conditions for a universal benchmark. The model has shown strong performance across a wide range of datasets.

Publication date: 4 Oct 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2309.17384