Radiated Noise Suppression for Electrolarynx Speech Based on Multiband Time-Domain Amplitude Modulation

Abstract

Radiated noise severely degrades the electrolarynx (EL) speech. It cannot be thoroughly suppressed by conventional frequency-domain enhancement methods. In this paper, a new method, called multiband time-domain amplitude modulation (MTAM), is proposed to reduce the radiated noise of EL speech. In the proposed method, the speech components changing slowly that represent the radiated noise are removed by directly modulating the time-domain amplitudes in multiple frequency bands. The EL speech enhanced by the proposed MTAM and the conventional frequency-domain enhancement methods (spectral subtraction and Wiener filtering) are evaluated on both acoustic and perceptual characteristics. The acoustic analysis reveals that the MTAM not only can reduce the radiated noise more thoroughly but can also easily control the residual noise intensity by adjusting a modulation parameter λ. Moreover, the MTAM can avoid causing new artificial noise that cannot be avoided by the conventional frequency-domain enhancement methods. The perceptual analysis indicates that the MTAM also have better performance on increasing the acceptability and the consonant intelligibility of EL speech than spectral subtraction and Wiener filtering. These findings validate that the MTAM indeed works well in suppressing the radiated noise of EL speech and avoiding the artificial noise.