Noise Reduction within Network through Modification of LPC Parameters

Abstract:
Speech enhancement like noise reduction is usually done as pre-processing before speech encoding. Voice Quality Enhancement algorithms have recently appeared as a solution to improve speech quality within networks. They involve decoding the bitstream, performing the enhancement on the decoded signal and re-encoding the processed speech. This method is computationally expensive, implies additive delay and reduces speech quality because of codec tandeming. As an alternative approach, modification of the bitstream itself would definitely avoid tandeming. In this context, a new approach to reduce distortions due to the environmental background noise is proposed in this article through a modification of the Linear Prediction filter in noisy condition. This solution is compared with classical solution based on decoding, applying Wiener filtering in the frequency domain and re-encoding the signal.