The impact of speech detection errors on the noise reduction performance of multi-channel Wiener filtering

While the noise reduction performance of the Generalized Sidelobe Canceller (GSC) depends on the validity of a priori assumptions about the signal model, the recently developed multi-channel Wiener filter technique does not rely on any such a priori information. This provides a potential benefit of the latter over GSC. However, both techniques also rely on a speech detection algorithm. In this paper, we analyze the average effect of speech detection errors on the performance of the GSC and the multi-channel Wiener filtering technique both theoretically and experimentally. In the GSC case, it is the simultaneous presence of signal model errors and speech detection errors that affects performance. Theoretical analysis (for infinite filter lengths) shows that already in the presence of small signal model errors, the GSC is more affected by speech detection errors than the multi-channel Wiener filter, both with and without Adaptive Noise Canceller (ANC) postprocessing. Incorporating a constraint on the noise sensitivity of the GSC limits the drastic impact of speech detection errors at the expense of reduced noise reduction performance. It is shown that the multi-channel Wiener filter technique preserves its benefit over the GSC for a reasonable speech detection error rate of Y a ` c b or less, even when the GSC is supplied with a noise sensitivity constraint. Real data experiments confirm the theoretical results.