Perceptual Error Identification of Human and Synthesized Voices

Performance of Phonatory Deviation Diagrams in Synthesized Voice Analysis

“…Some criteria guarantee the external validity and consequently a possible transference of the results found here to the clinical context with respect to the relationship between the PDD and the auditory-perceptual evaluation. These include the use of a set of synthesized signals validated in previous studies [43, 44], which are considered suitable for simulating vocal deviations with an excellent level of naturalness and characteristics similar to the signals of dysphonic patients, as well as synthesized signals generated with a combination and gradation of acoustic parameters, representing a wide range of deviation (mild, moderate, and severe) and different types of voices (rough, breathy, and strained).…”

“…The signals produced by this synthesizer were previously evaluated for quality and naturalness [43, 44] and were considered to be excellent representations of similar healthy and deviated human vocal production that considered the different types and degrees of deviation.…”

“…For further details on the synthesizer and sample generation, please refer to Lucero et al [48] and Englert et al [43, 44]. …”

“…Breathiness was related to audible airflow in the vocal emission, and strain corresponded to perceived vocal effort. Furthermore, the rater was trained with anchor stimuli containing normal vocal emissions and with different degrees of deviations and predominantly rough, breathy, and strained voices, which were assessed and validated in a previous study [43, 44]. …”

“…Therefore, to overcome this limitation, using validated and tested signals is recommended, where the relationship between the acoustic, physiological, and biomechanical parameters used to produce these signals and the auditory effect generated by the manipulation of these parameters is known [43]. Synthesized voices must have a correlation between acoustic and auditory-perceptual parameters [39], representing a model of what occurs with human signals [11], and should sound as natural as possible to listeners [44].…”