Abnormal sound detection by two microphones using virtual microphone technique

In this paper, we propose a new method of microphone array signal processing for detecting abnormal sounds that is applicable to monitoring elderly people at home and can be implemented on small equipment. This method consists of noise reduction based on a subspace method and sound activity detection (SAD), which is the same as voice activity detection using the signal power. The performance of noise reduction may degrade for underdetermined conditions (the number of microphones is less than that of sound sources). To resolve this issue, we previously proposed a technique of microphone array signal processing that introduced virtual microphones. In this method, signals of virtual microphones are interpolated with those of real ones. By using both real and virtual microphones, this noise reduction method can be applied for a critical / overdetermined condition. In this paper, we apply this method to the subspace method for the first time. After noise reduction, the abnormal sounds can be detected by the SAD method. We conducted an experiment and confirm that the proposed method is effective for detecting abnormal sounds in noisy environments and is robust to abnormal sound directions.