Contextual Relativity of Speech Recognition Cues
Martin C. Schultz · The Journal of the Acoustical Society of America · 1971
The use of distinctive features as cues to speech recognition should result in performances for signals that differ as a function of response alternatives. Recognition of a speech unit that shares many features of alternative responses must rely upon reception of the one or few uniquely held features. Recognition of a speech unit that shares few, if any, features will occur with reception of (almost) any of its features. A four-person talker-listener round robin generated nonsense syllables of a CV type using the consonant /t/ and two others, and the vowel /a/ and two others. Each of the nine syllables was delivered 100 times in random order in each set. Each listener responded with a choice for each signal plus a confidence rating on his decision. Additional sets were generated by retaining the consonants and the /a/ but using two different vowels, or by retaining the vowels and /t/ and using two different consonants. Other sets used the original elements but always placed an /r/ for a CrV syllable, and still others added a final /k/ for a CrVk syllable. Utilities of response choices were systematically varied and response confusions generated for each talker-listener dyad. Results on stability of recognition performance pertinent to the influence of context are discussed.