Speech Quality Assessment

This chapter provides an overview of the various methods and techniques used for assessment of speech quality. A summary is given of some of the most commonly used listening tests designed to obtain reliable ratings of the quality of processed speech from human listeners. Considerations for conducting successful subjective listening tests are given along with cautions that need to be exercised. While the listening tests are considered the gold standard in terms of assessment of speech quality, they can be costly and time consuming. For that reason, much research effort has been placed on devising objective measures that correlate highly with subjective rating scores. An overview of some of the most commonly used objective measures is provided along with a discussion on how well they correlate with subjective listening tests. The rapid increase in usage of speech processing algorithms in multi-media and telecommunications applications raises the need for speech quality evaluation. Accurate and reliable assessment of speech quality is thus becoming vital for the satisfaction of the end-user or customer of the deployed speech processing systems (e.g., cell phone, speech synthesis system, etc.). Assessment of speech quality can be done using subjective listening tests or using objective quality measures. Subjective evaluation involves comparisons of original and processed speech signals by a group of listeners who are asked to rate the quality of speech along a pre-determined scale. Objective evaluation involves a mathematical comparison of the original and processed speech signals. Objective measures quantify quality by measuring the numerical “distance” between the original and processed signals. Clearly, for the objective measure to be valid, it needs to correlate well with subjective listening tests, and for that reason, much research has been focused on developing objective measures that modeled various aspects of the auditory system. This Chapter provides an overview of the various subjective and objective measures proposed in the literature [1] [2, Ch. 10] for assessing the quality of processed speech. Quality is only one of many attributes of the speech signal. Intelligibility is a different attribute and the two are not equivalent. For that reason, different assessment methods are used to evaluate quality and intelligibility of processed speech. Quality is highly subjective in nature and it is difficult to evaluate reliably. This is partly because individual listeners have different internal standards of what constitutes “good” or “poor” quality, resulting in large variability in rating scores