Nonstationary Noise Reduction in Low SNR Speech Signals with Wavelet Coefficient Feature

摘要

Generally, the real-time environments with nonstationary noises and speech enhancement under low signal-to-noise ratio (SNR) remains a challenging task. Consequently, this article proposes a supervised model for speech enhancement by using the wavelet coefficient as an input feature. The performance of speech enhancement improves with the input feature selection (i.e., Wavelet coefficient). Experimental results are observed in terms of speech quality and intelligibility. Quality is evaluated with the perceptual evaluation of speech quality (PESQ). The intelligibility of speech is measured with the score of short term objective intelligibility (STOI). Speech enhancement results in a known and unknown non-stationary noise environments that are compared with conventional and supervised speech enhancement models. The result shows that the proposed feature has an improved intelligibility in diverse environmental conditions.