Audio Signal Processing and Coding

12.6 SUBJECTIVE EVALUATIONS USING PERCEPTUAL MEASUREMENT SYSTEMS

Even well-designed and carefully controlled experiments [ITUR94b] are susceptible to problems with reliability, repeatability, site-dependence, and subject-dependence [Grew91]. Moreover, subjective tests are difficult, time consuming, tedious, and expensive. These circumstances have inspired researchers to seek automatic perceptual measurement techniques that can evaluate coding distortion on a subjective impairment scale, with the ultimate objective being the replacement of human test subjects. Ideally, a perceptual measurement system should answer the following questions:

• Is the codec transparent for all sources?
• If so, what is the “coding margin,” or the degree of distortion inaudibility?
• If not, how disturbing are the artifacts on a meaningful subjective scale?

Table 12.3. Comparison of standardized 5.1-channel algorithms.

In other words, the system should quantify overall subjective quality, provide a distance measure for the coding margin (“perceptual headroom”), and quantify the audibility of unmasked distortion. Many researchers have proposed measurement systems that attempt to answer these questions. The resulting system architectures can be broadly categorized into two classes (Figure 12.2), namely, systems that compare internal auditory representations (CIR), and systems that perform noise signal evaluation ...