Audio Signal Processing and Coding

12.9 RESEARCH DIRECTIONS FOR PERCEPTUAL CODEC QUALITY MEASURES

This chapter has provided a perspective on current performance evaluation techniques for perceptual audio codecs, including both subjective and objective quality measures. At present, the subjective listening test remains the most reliable technique for measurement of the small impairments associated with high-quality codecs, and a formal test methodology optimized for maximum reliability has been adopted as an international standard, the ITU-R BS.1116 [ITUR94b]. Nevertheless, even the results of BS.1116-compliant subjective tests continue to exhibit site and subject dependencies. Future research will continue to seek new methods for improving the reliability of subjective test outcomes. For example, Lucent Technologies sponsored an investigation at Moulton Laboratories [Moul98a][Moul98b] into the effectiveness of multifacet Rasch models [Lina94] for improved reliability of subjective listening tests on high-quality audio codecs. Developed initially in the context of intelligence testing, the Rasch model [Rasc80] is a statistical analysis technique designed to remove the effects of local disturbances on test outcomes. This is accomplished by computing an expected data value for each data point that can be compared with the actual collected value to determine whether or not the data point is consistent with the demands of reproducible measurement. The Rasch methodology involves an iterative data pruning process of suspending ...