Analysis of Subjective Data from the MPEG Unified Speech and Audio Coding Call for Proposals
In October 2007 ISO/IEC MPEG issued a Call for Unified Speech and Audio Coding. Eleven technologies were proposed and each was evaluated at nine operating points (i.e. bit rate for either mono or stereo signals) using the MUSHRA test methodology. The test was conducted at seven sites comprising a total of 64,500 individual subjective scores. The paper contrasts a simple 95% confidence interval on the mean scores and an analysis of variance (ANOVA) model for the set of data. A deeper analysis validated the assumption of Gaussian distribution for the model errors and the ANOVA methodology gave more that 25% greater power than the simple grand mean analysis. In other words, the ANOVA reduced the 95% confidence interval on the grand mean by 25% or, conversely, would have permitted fewer listeners in the testing effort in order to get the same results as the simple grand mean analysis.
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 temporarily free for AES members.