Speech quality assessment using EEG signals

Abstract

The listener experience in modern audio communication systems acts as a key indicator for the entire system quality. Consequently, speech quality assessment attracts great interest from both industry and academia. Common methods for speech quality assessment are either subjective or based on subjective experiments and therefore limited in their ability to produce unbiased results. In this paper, we introduce a new EEG-based distortion measure (EBDM) for speech quality assessment. The EEG signals are represented by multi-channel autoregressive parameters. These parameters are then used, in coordination with Dynamic Time Warping algorithm to compare EEG responses of two signals with different speech quality. The method demonstrates promising preliminary indications for the possibility of using EEG signals as an objective basis for assessing speech quality degradation levels. We believe that further improvements will allow such EEG-based methods to be competitive with standard ITU-T recommended quality testing.