Comparative Analysis of Compressive Sensing Methods for Speech Enhancement in Hearing Aid Applications

Abstract

Enhancing the speech corrupted by real-world noises is a challenge for hearing aid applications. To address it, various filters have been designed with a goal to improve the quality and intelligibility of speech. Compressing sensing (CS) is found to be a promising technique for enhancing speech in various speech processing applications. However, the suitability of CS is limited for hearing aid applications due to higher reconstruction delay. Towards this end, a machine learning-based optimization to CS has been proposed in this work. The proposed solution is a detailed and comparative analysis of various optimizations to CS against the proposed machine learning-based optimization is presented. The performance is compared in terms of standard speech quality metrics and speech reconstruction time. The machine learning-based optimization to CS reconstruction achieves higher performance in both speech quality metric and speech reconstruction time.