Speech Signal Enhancement Using Spectral Subtraction Method

This MATLAB implementation performs speech signal enhancement using the spectral subtraction method. Through comprehensive testing and evaluation, we have confirmed its excellent performance in improving speech quality. Spectral subtraction is a widely-used speech signal processing technique that enhances clarity and audibility by subtracting estimated noise components from the speech signal. The algorithm typically involves: computing the short-time Fourier transform (STFT) of the noisy signal, estimating the noise spectrum during non-speech segments, subtracting the noise spectrum from the overall spectrum, and reconstructing the enhanced signal through inverse STFT. This method finds extensive applications in speech recognition systems, speech synthesis technologies, and communication fields. The implementation includes key functions for frame blocking, windowing, spectral analysis, and noise estimation, making spectral subtraction an effective approach for improving speech transmission quality and reliability. The code features adjustable parameters for noise estimation windows and subtraction factors, allowing customization for different acoustic environments.