Two-channel Coherence-Based Own Voice Detection for Privacy-aware Longterm Acoustic Measurements

Abstract:
Based on a two-channel behind-the-ear microphone attached to a smartphone we developed an application that can be used for long-term recordings of parameters without disturbing the privacy of the user or bystanders. In order to use the recorded parameter to compute acoustical meaningful parameters it is necessary to remove the parts which contain the user’s voice. In this contribution a coherence-based method is proposed to detect the own voice without access to the original audio data. In comparison to a known method that needs access to the audio signal directly the new method gives similar results in terms of robustness and accuracy.