Enhancement of speech signal corrupted by high acoustic noise

Additive broadband noise is a common source of corruption in speech signals and often limits the effective operation of speech communication systems. This paper presents a new spectral-subtraction-based speech enhancement algorithm which exploits the correlation property of speech signal, the masking characteristic of the human ear, the speech redundancy property, and the frame-to-frame randomness of the noise that is present in the signal for the enhancement. Informal listening tests show that the proposed algorithm greatly improved the quality of the noisy speech even when the speech signal is corrupted by high acoustic noise. As the algorithm has low complexity and uses only a single input, it is well suited to a wide range of speech processing applications.<>