Coherent Modulation Comb Filtering for Enhancing Speech in Wind Noise

Enhancing single-channel speech corrupted by wind noise has proven to be a difficult topic due to the complex characteristics of wind noise. Methods that assume a stationary or quasistationary noise source are ineffective against wind noise due to its nonstationarity and unpredictability. In contrast, the new method proposed works by finding the elements of the signal that are speech-like and suppressing the noise. This method takes advantage of the harmonic nature of speech by using a coherent modulation comb filter. Traditionally, very high-order IIR filters have potentially crippling stability constraints, but the proposed method bypasses these constraints by using coherent demodulation to filter harmonic subsets with lower-order filters. Potential applications for this research include mobile phones, audio production software, and as a front-end for automatic speech recognition (ASR) systems.