Inplace Cepstral Speech Enhancement System for the ICASSP 2023 Clarity Challenge

Abstract

This report summarizes our system submission to the ICASSP 2023 Clarity Challenge. The goal of the challenge is to estimate clean binaural speech signals within a 5 ms system delay. Our submitted Inplace Cepstral Speech Enhancement (ICSE) system features the following aspects: First, we developed a low-latency short-time Fourier transform (LL-STFT) analysis and synthesis strategy for a neural network-based speech enhancement algorithm in the time-frequency domain. Second, we designed an end-to-end Inplace Cepstral Speech Enhancement neural network that achieves good spatial resolution in an inplace speech enhancement framework. We also combined the cepstrum space speech enhancement with the TF-domain speech enhancement in the proposed system. Finally, we employed a speech model-based perceptual loss to improve speech intelligibility and quality. The experimental results show that the proposed system significantly outperforms the baseline system and ranked among the top five systems.