Article

A Robust Acoustic Head Orientation Estimation Algorithm Implementation - Evaluation for Real ICC Scenarios

* Presenting author
Day / Time: 21.03.2019, 08:40-09:00
Room: Saal 5-a
Typ: Vortrag (strukturierte Sitzung)
Abstract: In order to improve speech intelligibility and communication quality among the passengers in a car driven at high or even moderate speed, so-called In-car communication (ICC) systems are recently used. ICC systems usually work by recording, processing, and playing-back the desired speech signal of the talking passenger over loudspeakers located close to the passengers. Due to the acoustic directionality of human head, various signal processing technique are required to reduce the degraded intelligibility and distortion in the speech signal which could arise when the talking passengers turn their head. In our last contribution, we had presented the first investigations towards a robust acoustic head orientation estimation of the speaking passenger within a simulated noisy car compartment. We were able to estimate the head orientation horizontal angle (azimuth) precisely with less than 2 degree mean magnitude error even at the worst case (- 5 dB scenario). However, in this contribution we are aiming on implementing a more robust head orientation estimation algorithm with a real-time base estimation and evaluating its performance in a real car environment with different test case scenarios. In addition, we are aiming to extend our algorithm to estimate the vertical angle of the speaker head (elevation)