IVA-based Real-time Source Extraction of a Moving Speaker

Johannes Zeitler and Andreas Brendel
Friedrich-Alexander-Universität Erlangen-Nürnberg

In this demo, we showcase the extraction of a moving acoustic source observed within a mixture of undesired sources and background noise. The extraction framework is using a single small microphone array and is based on a recently developed IVA algorithm exploiting spatial priors formed from continuously estimated DOAs. The demonstrated algorithm is computationally efficient and can operate on low-cost embedded hardware, such as a Raspberry Pi, in realtime. Visitors can experience the performance of the proposed method in a real-world scenario with human speakers.