Improving Speech Recognition Performance using Spectral Subtraction with Artificial Neural Network
J. Umamaheswari, A. Akila · SSRN Electronic Journal · 2018
This study proposes a new technique for speech enhancement using Time-Delay Neural Network Spectral Subtraction (TDNN-SS) in the presence of background noises. Spectral subtraction suppresses the noise in the speech process, and ANN is used for training and testing the speech data sets. The proposed TDNNSS is capable of enhancing the speech signal comparing with the existing methods. This proposed TDNSS can reduce the error rate in the speech recognition. This method also improves the quality of the signal by reducing the noise in the background signal. Further, this TDNSS based approach is validated with existing speech recognition methods.