Speech Intelligibility Improvement Using the Constraints on Speech Distortion and Noise Over-estimation

Existing speech enhancement methods can improve speech quality but not speech intelligibility, especially in low SNR conditions. To solve this problem, an algorithm for improving speech intelligibility using the Constraints on Speech Distortion and Noise Over-estimation (CSDNO) is proposed in this paper. Based on the fact that the attenuation distortion and amplification distortion have different impacts on the speech intelligibility, the gain function and noise estimation method used in conventional algorithm are modified in this paper. The performance of the proposed method has been evaluated by Diagnosis Rhyme Test (DRT), fractional Articulation Index (fAI) and frequency-weighted SNR segmental (fwSNRseg) for three types of noises. The experimental results show that, with smaller speech distortion, the proposed method can improve the intelligibility of the enhanced speech in comparison with the reference method.