Speech protected noise cancellation system in noise dominated environments

Abstract

People working in noisy environments should wear hearing protective devices (HPD) to prevent noise-induced hearing loss. Since traditional HPDs suppress all sounds in the environment, users are exposed to dangers as they are unaware of verbal instructions and warnings. Thus, HPDs need speech protected noise cancellation systems that isolate and enhance speech while reducing harmful background noise. In this paper, a speech protected noise cancellation system consisting of dual voice activity detection (VAD), convolutive blind source separation (CBSS), and noise cancellation blocks, is presented. In the VAD block, to find the feature that best detects speech in a noisy environment, different feature extraction methods are implemented, and also a deep neural network (DNN), that extracts features from raw data is employed. The performances of most speech protected noise cancellation systems in the literature have been calculated at high signal-to-noise ratios (SNRs) that are unrealistic for working environments and only in the presence of speech situations. In this study, the performance of the proposed system is compared with two different speech protected noise cancellation systems in the presence and absence of speech, at different SNRs from low to high, and in two different microphone configurations. The results indicate that, while the proposed system protects the speech successfully in the presence of speech, its noise suppression performance is better than the other tested systems in the absence of speech.