29/11/2023 Combining Distributed Microphones in Smart (IoT) Environments for Acoustics Source Separation

In our everyday lives, we encounter a lot of devices which carry at least one microphone with them: think of smart phones, smart watches or laptops. Combining these microphones could be of great benefit for improving audio quality in for example hybrid meeting settings or for hearing aid users.

However, the location of these microphones with respect to each other and the active speakers are unknown. Finding the best solution to this problem is exactly one of the research questions raised in the IMEC.ICON project ‘BLE2AV’ where we want to use bluetooth low energy (BLE) for (distributed) audio and voice (AV) applications.

At IDLab, we have developed a method, based on speaker verification embeddings, that can cluster microphones with respect to relevant speech sources. These clustered microphones can subsequently be used for tasks like speech enhancement or speaker separation.

In the past, we have tested the approach with simulated data. Thanks to HomeLab and the equipment of Televic, partner in the BLE2AV project, we were able to make recordings that can better approximate real live situations to check the performance and find possible mismatches not yet accounted for.

The image shows three speakers that were put in the room and 18 different individual microphones, distributed in the room. Televic conferencing system was used, which can connect all the individual microphones to one central recording device.