Channel prediction-based noise reduction algorithm for dual-microphone mobile phones

In this paper, a noise reduction algorithm for dual-microphone mobile phones is proposed, in which a frequency-domain channel prediction filter combined with a variable step-size scheme is utilized to adaptively estimate and suppress the coherent noise component in low-frequency region. To suppress the incoherent noise component, the proposed algorithm estimates a noise power spectral density (PSD) from the error signal of the channel prediction filter using a voice activity detection (VAD) scheme based on a SNR ratio between the two microphone signals. Computer simulations were performed using synthesized and recorded noises, and the results confirmed that the proposed algorithm obtains more accurate noise PSD especially in low-frequency band and better overall speech quality compared with the conventional algorithms.