Musical noise reduction in speech using two-dimensional spectrogram enhancement

This paper investigates the problem of "musical noise" and proposes a new algorithm to reduce it. Musical noise occurs in most of the spectral-estimation-based algorithms, such as spectral subtraction and minimum mean-square error short-time spectral amplitude estimator (MMSE-STSA). To reduce this type of noise, a novel algorithm, which is called two-dimensional spectogram enhancement, is proposed. A speech enhancement scheme is implemented by combining the proposed algorithm with the MMSE-STSA method. Spectogram comparisons show that with the proposed scheme, musical noise is effectively reduced with reference to MMSE-STSA. SNR and PESQ evaluations show that the proposed method is superior to MMSE-STSA and spectral subtraction with auditory masking method.