| The purpose of this study was to evaluate the production and recognition of speech spoken in four environments: quiet, wideband noise (WBN), speech spectrum noise (SSN), and meaningful multitalker competitor (MMC). Specifically, differences in recognition for speech spoken in quiet (SSIQ) and speech spoken in a competitor (SSIC) were measured. Speech samples were gathered from five female speakers reading Speech Perception in Noise-Low Predictability (SPIN-PL) sentences (Kalikow, Stevens, & Elliot, 1977). Mean vocal level (dB SPL), spectral tilt (dB SPL/kHz), target word duration (ms), and speech rate (ms/syllable) were measured for the vowels (ae) and (a) produced in each speaking environment. Mean values for speech spoken in WBN, SSN, and MMC were compared to SSIQ. Additional comparisons, WBN vs. SSN and SSN vs. MMC were also performed to determine if the spectral and semantic content of these competitors influenced speech production. Significant increases in vocal level, spectral tilt, and target word duration were found in each of the competitive speaking environments. No significant differences in speech rate were found for SSIQ and SSIC. Comparison of the speech samples produced in each competitor (WBN vs. SSN and SSN vs. MMC) revealed no significant differences in vocal level, spectral tilt, word duration, or speech rate. These findings suggest that speech production was significantly influenced by the presence of a competitor but content of the competitors did not influence production. Percent correct recognition scores were gathered for selected speech samples produced in each speaking environment to determine the differences in recognition performance for SSIQ and SSIC. In separate conditions, recognition scores were gathered at several speech-to-noise levels, first with vocal level differences between the speech samples preserved, and second, with the vocal level differences removed. Recognition scores for SSIC were found to be significantly higher than SSIQ when vocal level differences between the two types of speech were preserved. When vocal level differences were removed, recognition scores for SSIC were again significantly higher than for SSIQ suggesting that properties other than vocal level were responsible for the better recognition performance found for each speech sample produced in the competitors. |
