Scalable Perceptual Metric for Evaluating Audio Impairment

Masters thesis abstract:

ITU-R BS.1387-1 presents a method for objective measurement of perceived audio quality know as PEAQ (Perceptual Evaluation of Audio Quality). The PEAQ has been designed to perform optimally for evaluating high quality audio. In this thesis, we show that the PEAQ advanced version performs poorly for evaluating the quality of audio that is highly impaired when compared to the Energy Equalization approach (EEA). We also introduce in this work a new metric that uses six Model Output Variables (MOVs) – five MOVs from the original PEAQ advanced version and one MOV from the EEA. These MOVs are mapped to a single quality measure using an optimized single-layer neural network. This metric performs better than both EEA and PEAQ advanced version for measuring audio quality over a wide range of impairment. Furthermore, by using the bitrate information of the encoded audio signal, the performance of the proposed approach is shown to further improve. The final result of our research is an audio quality metric capable of accurately predicting perceptual quality over a wide range of audio impairment – i.e., a scalable metric.

Note:
Advisor: Charles Creusere.

Email any questions/comments/correction: rahulv AT u DOT washington.edu
Last updated: 12/02/2006