The goal of the proposed research is to develop procedures for predicting perceptual confusions of speech sounds in noise by integrating knowledge of the acoustic properties of the speech signal with that of the properties of the human auditory system. This goal guided the design of the set of experiments proposed for the study. A long-term objective is to contribute towards a more complete and quantitative theory of speech perception. Although noise is very frequently the limiting factor in normal communication, most previous studies that examined the perceptual importance of acoustic cues in signaling phonetic contrasts have been based on experiments conducted in quiet. In the current proposed study, the perception of some consonant-vowel (CV) syllables in noise is considered. The methodology of the study has three component: 1) quantifying acoustic correlates of some phonological features in naturally-spoken utterances, 2) developing a metric to predict the level and spectrum of the noise which will mask these acoustic correlates, and 3) performing a series of perceptual experiments to evaluate the theoretical predictions. The focus will be on the plosives /b,d,g,p,t,k/, the nasals /m,n/ and the fricatives /s,z/ in syllable-initial position with the vowels /alpha/ and /epsilon/. The metric for predicting masking is based on a combination of theoretical and empirical results; the theoretical part is based mainly on auditory masking theory and the empirical part is based on the results of a pilot study and six proposed experiments. The perceptual study consists of 12 experiments which examine the perception of synthetic an natural CV syllables in noise. In each experiment, the perceptual importance of an acoustic cue known to signal a particular phonetic contrast will be studied by selectively masking the cue with a noise masker. Based on the masking metric and the results of the perceptual experiments, a model of consonant perception in noise is proposed. In the final phase of the project, all of the above mentioned consonants will be presented jointly to subjects in two identification-in-noise experiments. The masker in these experiments is either white noise or speech-spectrum noise. Resulting confusion matrices will be compared with model predictions.
| Alwan, Abeer; Jiang, Jintao; Chen, Willa (2011) Perception of Place of Articulation for Plosives and Fricatives in Noise. Speech Commun 53:195-209 |
| Hant, J J; Strope, B P; Alwan, A A (1998) Variable-duration notched-noise experiments in a broadband noise context. J Acoust Soc Am 104:2451-6 |
| Hant, J J; Strope, B P; Alwan, A A (1997) A psychoacoustic model for the noise masking of plosive bursts. J Acoust Soc Am 101:2789-802 |
