A background noise reduction technique based on sinusoidal speech coding systems

This paper describes a new technique to reduce background noise in speech signals in conjunction with a sinusoidal based vocoder in order to improve the quality and intelligibility of processed speech. In a mobile communications environment, speech signals are degraded by additive random noise. Randomness of the noise, which is often described in terms of its first and second order statistics, makes it difficult to remove without introducing background artifacts. This is particularly true for lower signal to background noise ratios. The technique described in this paper deals with the noise reduction of the vocoded speech signals without any knowledge of the signal to background noise ratio. In general, low bit rate speech coding systems do not have their own mechanism to reduce the background noise from the speech signal. This is due to the complexity of the speech signal and limitations in the scope of many speech coding systems. As a consequence, most speech enhancement systems to date have attempted to process the speech waveform directly and independently from the speech coding system, before the encoding of speech signals. In this paper, we propose a new noise reduction technique which is based upon the harmonic excitation linear predictive coding (HE-LPC) and can be applied to and integrated with any sinusoidal based speech coding algorithm.