Abstract:
To improve the accuracy of noise estimate and the performance of speech enhancement algorithms, this paper proposes a noise classification based speech enhancement approach by using the improved minima controlled recursive averaging (IMCRA) algorithm. In this proposed algorithm, the type of noisy speech is firstly determined, which will be utilized to select the optimal parameters IMCRA to estimate the noise spectrum. Finally, the enhanced speech is calculated by the optimally modified log spectral amplitude speech estimator. Compared with the traditional IMCRA algorithm, the proposed approach can attain better performance in both the restoration of the speech signal and the suppression of background noise.