Audio Expansion Techniques For Real-Time Communication And Intelligibility Enhancement

Clear speech is a long-recognized and long-studied phenomenon which arises when a talker produces speech for the benefit of a listener distressed by, e.g., hearing difficulties, noisy environments, lack of language skill, and other similar conditions. Such speech is characterized by a number of differences from casual or conversational speech, but the most prominent, easily explained, and easily detected by untrained listeners, is a reduction in speech rate. Informally, people engaging in clear speech “speak slower.” Clear speech, when produced spontaneously or by instruction, is known to benefit listeners as they extract meaning and intelligibility from that speech; however, the degree of intelligibility boost is also known to be highly idiosyncratic. Moreover, it is known to be idiosyncratic according to both the talker (some talkers provide more benefit than others) and the listener (some listeners derive more benefit than others.) Prior studies have tried to replicate clear speech, or its benefits, by artificially slowing down (or “expanding”) speech, with mixed results, but failed to address the idiosyncrasies noted above, and always in carefully pre-scripted situations. This dissertation investigates whether ever more powerful personal computers can enable tools allowing for this slowing of speech in real time (i.e., in conversational rather than laboratory environments) and whether placing control of this audio expansion in the hands of the users themselves can address and alleviate these idiosyncrasies. The results are mixed: Subjects do respond to increasing auditory distress by using more expansion and respond favorably to the technique in survey questions; however, no improvement in intelligibility is demonstrated.