Subjective intelligibility tests make use of a speaker and a listener panel. The goal is to derive a score for correct received speech fragments. Various types of tests are developed to get a measure for the performance of the communication channel under test. Different types of communication channels require a different test, e.g. an auditorium differs substantially from a vocal alarm system or just announcements.
Tests can be ordered by the type of speech: sentences, meaningful words, nonsense words, rhyme-words or a limited vocabulary as digits and the spell-alphabet.
A frequently used measuring method for obtaining word and phoneme scores is based on nonsense words of the CVC-word type (Consonant-Vowel-Consonant). This type of test was used for our studies on band-pass limiting and noise masking and various communication channels.
The effect of several variables on the final score, such as speaker, listener, and speaker gender can easily be analysed. Also, the relation between the CVC-score and the individual phoneme score (initial consonants, vowels, and final consonants), and phoneme-group score (fricatives, plosives, vowel-like consonants, and vowels) can be analysed. For analysis of the results with respect to individual phonemes or phonetic grouping a principal-component analysis or multi-dimensional scaling techniques can be used. It was found that at the phoneme level, four groups can be identified with a fairly similar response for various transmission conditions.
The individual scores for phoneme groups are useful as a diagnostic tool for improving communication systems and to improve predictive intelligibility measurements.
The relation between sentence intelligibility and the CVC-word score was also studied. For this study the Speech Reception Threshold (SRT) was used. This method tunes a condition to a 50% sentence intelligibility level by adding noise at a required signal-to-noise ratio. Beside the usual 50% sentence intelligibility level, the 25% and the 75% sentence intelligibility level could also be obtained from the data. A good relation could be found with the CVC-word score, as well as with combinations of some phoneme-group scores.
See also: Subjective-Intelligibility-2014-V2