For blind source separation (BSS) based applications where the aim is the reproduction of the separated signals, the perceived quality of the produced audio signals is an important key factor to rate these systems. In this paper several signal-derived features are compared to assess their relevance in reflecting the perceived audio quality of BSS signals. The most relevant features are combined in a multiple linear regression model to predict the perceptual quality. In order to cover a large variety of source signals and different algorithms, the reference ratings are obtained from extensive listening tests rating the BSS algorithms that participated in the Stereo Source Separation Campaigns 2007 (SASSEC) and 2008 (SiSEC). Results are presented for predicting the perceived quality of SiSEC items based on a model that was calibrated using SASSEC material.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.