Failure of formant bandwidth narrowing to improve speech reception in sensorineural impairment

Abstract

Poor auditory frequency resolution resulting in a “blurred” representation of the short-term speech spectrum may underly the particular difficulty experienced by sensorineurally impaired listeners in distinguishing consonants produced at different places of articulation. To determine whether compensation might be achieved by enhancing the distinctiveness of formant peaks, whispered exemplars of “bet,” “debt,” “get,” “bib,” “bid,” and “big” were synthesized with formant bandwidths varying between unnatural extremes. Psychoacoustical measures of frequency resolution in a combined group of 12 normal and 12 sensorineurally impaired listeners correlated significantly with accuracy of identifying these words although correlations within each group were not systematic. Independently, the psychoacoustical and speech identification measures each suggested that frequency resolution in the impaired group was about four times poorer than normal. The normals' speech-identification accuracy was near 90% at narrow and normal bandwidths, and deteriorated as bandwidth increased. Impaired listeners scored near 60% at normal bandwidths, and also performed worse as bandwidth increased, but contrary to expectations did not improve as bandwidth was narrowed. Possibly, unnaturalness outweighed any advantages of improved formant resolution. Alternatively, little may be gained by making formant bandwidths narrower than the widths of listeners' auditory filters. [Work supported by MRC and by SRC U.K.]