Joint enhancement and coding of speech by incorporating Wiener filtering in a CELP codec

The performance of speech communication applications in the field of mobile devices is often hampered by background noises and distortions. Therefore, noise attenuation methods are commonly used as a pre-processing method, cascaded with the speech-codec. We demonstrate that the performance of such combinations of speech enhancement and coding methods can be improved by joining the two methods into a single block. The proposed method is based on incorporating Wiener filtering into the objective function used for optimization of the quantization in code excited linear prediction (CELP)-based codecs. The benefits are that 1) the non-linear components of CELP codecs, including quantization and error feedback, are taken into account in the joint minimization function thereby improving quality and 2) by merging blocks both delay and computational complexity can be minimized. Our experiments demonstrate that the proposed joint enhancement and coding approach consistently improves subjective and objective quality. The proposed method is compatible with any CELP-based codecs without changing the bit-stream, whereby it can be readily applied in mobile phones or speech communication devices applying the concepts of CELP codecs for improving perceptual quality in adverse conditions.