Trends in adaptive MISO system identification for multichannel audio reproduction and speech communication

Abstract

Online identification of multiple-input/single-output (MISO) acoustic systems is one of the long-standing and continuing challenges in multichannel speech and audio applications. Fast and robust estimation of the impulse response of an acoustic system is a key requirement for several adaptive solutions in time-varying scenarios, such as stereophonic acoustic echo cancellation, room equalization, or crosstalk cancellation. The inevitable presence of cross-correlated loudspeaker signals that is implied by multichannel applications, however, entails the well-known non-uniqueness problem of MISO system identification. Apart from this fundamental issue, a more practical problem already consists in the lack of techniques to evaluate the estimated impulse responses properly. Since well-established measures are often not capable of accounting for all aspects of online MISO system identification, we revert to the recently proposed spectral-importance weighted misalignment (SIWM) to assess MISO identification. In this contribution, we review SIWM and its relation to well-established evaluation tools. On this basis, we provide an insight into the problem of MISO system identification in applications driven by real stereo data. We also analyze and compare a traditional and a very recent approach to deal with the non-uniqueness problem.