Colloquium - Details

Master-Vortrag: Speech Enhancement for Active Occlusion Cancellation in Noisy Environments

Frederik Blang
Freitag, 24. September 2021
14:00 Uhr
virtueller Konferenzraum

The natural perception of one's own voice while wearing a headset is increasingly in the focus of customers and manufacturers. Due to the occlusion effect which occurs when the ear canal is blocked, body-conducted sound is amplified, resulting in an unpleasant sound of the own voice. It becomes difficult to control the volume of one's own voice leading to earlier vocal fatigue. Active Occlusion Cancellation (AOC) aims to suppress this effect and can be combined with a hear-through system for a natural own-voice perception. In noisy environments, for example call centers, it is desirable to only equalize the own voice. Some business headsets already provide a sidetone which uses a boom microphone to record the own voice, however this signal often isn't equalized.

This thesis develops a real-time system which provides a natural own-voice perception for the user of a headset using the boom microphone. Besides the implementation of a filter for the direct sound, also the room reverberation of one's own voice is estimated using convolution. Algorithmic room simulation on a DSP is considered, but ultimately rejected due to memory card. The low-latency filtering of the direct sound is performed on the DSP, while the longer room filters are run on the PC. Evaluations using simulations and measurements show the direct-path filter to perform well, while the room filters can't provide a natural estimation of the room sound.

back