Matthias Schrammen Knihy

Handsfree speech communication devices face low signal-to-noise ratios, necessitating effective front-end signal processing. This includes using a beamformer with known microphone positions, enhanced by the orthogonal geometric projection for low user effort. A novel change prediction technique allows a single acoustic echo canceller to replace multiple ones, while a dual-stage multi-channel Kalman algorithm addresses nonlinear effects in consumer devices. Performance is tested in realistic scenarios.