Restoration of Environmental Noise Corrupted Speech Using Wavelet Transform

Three methods are used for restoring environmenta1 noise-corrupted speeches.百le methods use a conventiona1 Wavelet Transform, a Wavelet Packet Transforn, and low-pass futers based on moving-average. The prominent result is obtained by using a conventiona1 Wavelet Transform_ In也lS paper, we proposed a new method of optimum decomposition of the speech using the Wavelet Transform. l.Introduction An environmental noise affects the understanding of a speech and the performance of a speech recognition system_ As a noise reduction method, the spectral subtraction method (SS) [1] has been utilized. 1n the method, a noise component is assumed to be stationary , and is estimated from the non-speech duration. Then the estimated noise is subtracted from the noise-corrupted speech. The method needs three proced町 es as describe next : (1) discrimination of a non-speech and a speech duration; (2) processing of a nonstationary noise; and (3) processing of a ー141 negative spec位um_ These procedures need a complicated processing. 1n this paper, the restoration method of the noise-corrupted speech using a Wavelet Transform and a Wavelet Packet Transform [2] is described_ Moreover, a method utilizing a low-pass filter based on a moving-average is a1so described for comparison. The procedure based on a non-sinusoidal function is suitable for processing nonstatonary signals_ The Wavelet Transform based on the nonsinusoidal function is suitable for such signals. Furthermore, we can avoid the negative spectrum by using the Wavelet Transform. The Wavelet Packet Transform has a merit of precise analysis of the signals in contrast to a conventional Wavelet Transform. However, we could not obtain excellent noise reduction results by using the Wavelet Packet Transform. The low-pass filter has sharp cutoff characteristics, and removes intrinsic properties of the speech, therefore, degraded speech was obtained. Among the three methods, the method based on a conventional Wavelet Transform brought the excellent result .