Open-set Coding Schemes for Speech and Environmental Sounds

In the present study, we report on a series of coding schemes to classify errors in open-set recognition of speech and environmental sounds from previous perceptual learning experiments with cochlear implant simulations (Loebach and Pisoni, 2007; in press). Open-set responses to MRT and PB words were coded for place of articulation, manner of articulation and voicing errors in the identification of word initial and word final consonants. Open-set responses to meaningful and semantically anomalous Harvard sentences were coded for phonemic, lexical, and thematic errors in keywords. Open-set responses to environmental sounds were coded for errors in identifying the agent, action, and rhythm of the sounds. Overall, our coding scheme provided a more accurate assessment of performance producing higher percent correct recognition scores for the isolated words and environmental sounds than absolute coding schemes that simply identified entire words as correct or incorrect. Although time intensive, these coding schemes revealed perceptual elements with which subjects were having difficulty that were not apparent from the absolute coding scheme. The utility of open-set coding schemes is discussed for perceptual experiments with cochlear implant users, who often must make verbal responses to stimuli.