Spectral Gating for Noise Reduction in Speech Stress Recognition System

In recent years, speech-based subconscious emotion (stress) recognition systems have been widely explored. System performance decreases. Therefore, this paper proposes an emotional stress recognition system that is robust to noise by applying a spectrum gating algorithm before feature extraction. Spectrum gates are able to reduce noise at predetermined frequencies thereby improving signal quality. The Mel-Frequency Cepstral Coefficient (MFCC) technique and the Convolutional Neural Network (CNN) method are used which support the Spectral Gating algorithm to adapt more efficiently to new data. The proposed system is embedded in a smartphonebased application to increase the flexibility of the measuring instrument. The effectiveness of the proposed system reaches 60% in term of accuracy. By involving speech-based stress emotion measurements and noise reduction technology, this work contributes to the development of more effective and reliable systems in supporting mobile monitoring of mental health problems.