Beschreibung:
An estimation of the power spectral density (PSD) of noise is a crucial part to retrieve speech in a noisy environment. A novel estimation method for non-white noise of noisy speech on the basis of a generalized Gamma distribution is proposed. Because of highly non-stationary nature of speech, its probability density function (PDF) is difficult to derive using any modeling technique, while a segmental noise is more stationary and can be fitted more accurately by a generalized Gamma PDF, which is a natural extension of the Gaussian modeling of a non-white components distribution. In the experiment, different types of non-white noises are added to the clean speech signal at different SNRs to study the estimation of noise using different types of PDF. It is found that non-white noise spectrums fit more accurately on the generalized Gamma PDF with adaptive parameters instead of a Gaussian distribution function. The reported generalized Gamma PDF model shows the best performance to estimate the noise spectral amplitudes as compared with Minimum Statistics (MS), Speech absence Probability (SAP), and MMSE based PSD estimation methods. The performance of the proposed noise estimation is good when it is integrated with the speech enhancement technique as demonstrated by both the subjective and objective measures.