Dual-Mic Speech Enhancement Based on TF-GSC with Leakage Suppression and Signal Recovery

The transfer function-generalized sidelobe canceller (TF-GSC) is one of the most popular structures for adaptive beamformer used in multi-channel speech enhancement. Although TF-GSC has shown decent performance, a certain amount steering error inevitable, which causes leakage components through blocking matrix (BM) and distortion fixed (FBF) output. In this paper, we propose to suppress leaked signal output BM restore desired FBF TF-GSC. To reduce risk attenuating noise (ANC), component suppressed by applying gain function similar square-root Wiener filter, assuming that portion should be into Additionally, attenuated adding some microphone back, depending on how signals are related outputs. experimental results showed proposed outperformed conventional terms perceptual evaluation quality (PESQ) scores under various conditions direction arrivals interfering sources.