Efficient speech enhancement using spectral subtraction for car hands-free applications

Over-subtraction of noise spectrum to enhance noisy speech has been proven an effective method. However, this method suffers from the tradeoff between speech distortion and musical noise artifacts. In this paper, we modify the original method to alleviate the tradeoff problem while retaining its simplicity. The proposed algorithm has been incorporated into the design of a car kit which is implemented on a low-cost fixed-point DSP with capabilities of voice dialing, echo cancellation and speech enhancement.