Error concealment for speech over noisy channels by using audio watermarking

Abstract

In this thesis, the author presents a novel error concealment method of speech for noisy channels using audio watermarking. The proposed error concealment algorithm abbreviated as ECAW is basically a LSB-based scheme and embeds the k-means clustering of (n)th frame into (n-1)st frame before audio encoding. After the transmission, the erroneous or lost parts of (n)th frame is concealed by the clustering information embedded into (n-1)st frame. Before sending speech signal into channel, ECAW's error concealment strategy divides speech signal into frames of 5 ms and executes k-means clustering with k=32. Clustering information of each frame is embedded into the previous frame. This intro- duces a constant delay of 5 ms to channel. Then, on the receiver side, received speech signal is broken into frames of 5 ms, too and k-means clustering with k=32 again is executed. It's assumed that error detection is provided externally and if any error is detected, ECAW's error concealment strategy replace the erroneous part with the corresponding part of the center of assigned k-means cluster. Experimental results prove the efficacy of the proposed method in reducing dis- tortions despite some restrictions.