Speech quality tests: analysis and review of literature

Technology Speech Science

Speech Quality Tests: An Analysis and Review of the Literature

The speech measurement technology to be evaluated in this research is a speech quality test. According to the research, a speech quality test is "a routine procedure for examining one or more empirically restrictive quality features of perceived speech with the aim of making a quantitative statement on those features" (). Essentially, people are constantly using undirected speech in ordinary conversations in order to exchange information through spoken sentences. Yet, when speech becomes directed by some third party or director, the speaker's speech may end up being modified by these directions. Here, the research suggests that "by using specific directives, the experimenter can, in directed communications, adjust the influence of each quality feature" (). A speech quality test can ascertain how biased perception may end up being in the case of directed speech. There are four primary types of measurement in two specified tests, including analytic / utilitarian and subject-oriented / object oriented (). Utilitarian tests utilize a scale rating from a single dimension perspective, whereas analytical tests utilize a multitude of scales in order to attempt to quantify the perceptual features. An object-oriented analysis of speech can then be performed, which would be based on processor conditions, while a subject-oriented test focuses more on the role of the subject in the perception of speech quality. Such tests can be given to either naive or expert test subjects, depending on the underlying goal of the research. Naive test subjects have not been trained in speech analysis and represent a broader scale of the larger public. Experts have more experience and thus can be used in more specified test contexts.

One article that used speech quality testing was written by Stan et al. (2011). This was a study conducted in order to test the efficiency and believability of synthetic speech. Perceptual tests were conducted using speech quality testing from an object-based utilitarian approach, as described above. The study used an HMM-based speech synthesizer to test on the quality of Romanian language with native Romanian speakers to act as test subjects. Researchers discovered that configuration choices, which alter the pitch and frequency of the synthetic speech, can greatly impact its perception by native language speakers. Essentially, in HMM-based speech synthesis, higher waveform frequencies actually showed improved speech recognition on behalf of the listeners. This research can then serve as a way to help improve modern designs of HMM speech synthesis software to be even more realistic and effective for native language speakers. Overall, this was an appropriate use of the testing in a more modern context.