Voice Activity Detection Exploiting PEVD

In this work, we exploit PEVD-based processing using multiple microphones for voice activity detection (VAD). In the first approach published in SSPD 2022, we use PEVD as a preprocessor to further improve the performance of conventional single channel VAD algorithms. The figure below summarizes the system design.

In the second approach published in IWAENC 2022, we proposed a PEVD-based target speaker (TS) voice activity detection (VAD) algorithm. The entry of the target speaker into the acoustic scene is reformulated as an anomaly detection problem. The figure below summarizes the TS-VAD approach.

Demo Pages

1. PEVD Multichannel Preprocessing for VAD

2. PEVD-based TS-VAD

Talk

1. SSPD2022