Double‐talk detector based on audio watermarking.

Abstract

The double‐talk detection (DTD) algorithm is a vital part of an acoustic echo cancellation system commonly used in the telecommunication applications. The role of the DTD algorithm is to stop the adaptation process of the adaptive filter used for the estimation of echo path response in the presence of double‐talk condition. The typically used DTD solutions are often a compromise between high efficiency and affordable computational cost. The algorithm introduced by the authors is based on a novel principle related to the so‐called semi‐fragile watermarking, making it possible to achieve satisfactory performance with limited processing burden. A hidden signature is embedded into the arriving far‐end speaker signal just before it is replayed to the near‐end speaker. Consequently, the presence of that signature is detected in the signal recorded by the near‐end speaker microphone. The appropriate choice of the employed watermarking technique causes the signature to become undetectable during the near‐end speaker’s talkspurts, precisely identifying the double‐talk periods. The use of echo‐hiding watermarking method allows fulfilling the algorithm requirements. The results of performed experiments prove the good performance of the proposed DTD solution. [Research funded by the Polish Ministry of Science and Higher Education within Grant No. PBZ‐MNiSW‐02‐II‐2007.]