Quantitative Investigation of Clear Speech: Identification of a Core-Set of Features

A number of studies in the literature have established a wide range of acoustic-phonetic modifications that typically characterize clear speech. Changes are observed in the temporal and spectral domains at the global level (e.g., decrease in speech rate, increase in the frequency and duration of pauses, in pitch range and mean fundamental frequency) and at the segmental level (e.g., more frequent releasing of word-final consonants, an increase in duration contrasts between the voice-onset-time of voiced and voiceless stops, an expanded vowel space). The long term objectives of this study are to identify a core-set of features to characterize clear speech and to develop an algorithm for the automated evaluation and classification of clear speech. To this purpose, the study is made up of the following steps: i) thorough review of the literature to identify a comprehensive list of clear speech features; ii) critical evaluation of the list of features to eliminate redundancy; iii) implementation of quantitative measures of the features identified in (ii) and preliminary characterization of their behavior in clear and degraded speech; iv) implementation of a comprehensive toolbox to extract the selected features from speech; and v) analysis of the sensitivity and accuracy of the different features and their combinations to characterize clear speech and identification of a core-set of features. The final outcome will be the availability of a toolbox for the automatic evaluation of speech clarity that could be used, for example, to assess clarity and quality of speech processed by hearing systems or implants.