Speech enhancement using modified IMCRA and OMLSA methods

In this paper we present a speech enhancement method in highly non-stationary noise environments based on modified Improved Minimal Controlled Recursive Averaging (IMCRA) method and Optimal Modified Minimum Mean-Square Error Log-Spectral Amplitude (OMLSA) method. The original OMLSA method, the spectral gain function, which minimizes the mean-square error of the log-spectral amplitude, is obtained as a weighted geometric mean of the hypothetical gain associated with the presence uncertainty. Whereas in IMCRA method, noise estimation is given by averaging past spectral value of noisy speech using a smoothing parameter that is adjusted by speech presence probability in frequency domain. A new method is proposed, in which the minimum spectral power value of noisy speech is adjusted by past speech presence probability. In addition, a noise estimation algorithm is proposed for highly non-stationary noise environment. The noise estimate is updated by averaging the noise spectral power estimate of IMCRA method with the past noise spectral power. Evaluations under various environment conditions, especially highly non-stationary noise environment, confirm that the modification of IMCRA and OMLSA method improved the speech quality.