Two-step noise reduction based on soft mask for robust speaker identification

This paper addresses the problem of speaker identification in noisy conditions. A two-step noise reduction algorithm based on soft mask and minimum mean square error short-time spectral amplitude estimator was proposed. It is used in the signal preprocessing stage for more robust speaker identification. The proposed algorithm was tested and compared with the existing noise reduction algorithms in the problem of speaker identification. Testing was carried out with two speech databases and some noise samples from the NOISEX-92 library. The advantage of the new noise reduction algorithm for some noise samples and signal-to-noise ratios was shown.