Research on Speech Recognition Methods Based on Spectral Subtraction and Improved BP Neural Network Algorithm

This paper presents a speech recognition method combining spectral subtraction with improved BP neural network algorithm. The speech samples containing unstable noise are collected in the real environment. Firstly, the noise from speech signals is removed using spectral subtraction method, and the speech feature parameters after noise removal are extracted. Then the speech is recognized effectively by the improved BP neural network algorithm. Finally, some comparative experiments are given to verify the effectiveness of the proposed method.