Individual and environment-related acoustic-phonetic strategies for communicating in adverse conditions

In many situations it is necessary to produce speech in ‘adverse conditions’: that is, conditions that make speech communication difficult. Research has demonstrated that speaker strategies, as described by a range of acoustic-phonetic measures, can vary both at the individual level and according to the environment, and are argued to facilitate communication. There has been debate as to the environmental specificity of these adaptations, and their effectiveness in overcoming communication difficulty. Furthermore, the manner and extent to which adaptation strategies differ between individuals is not yet well understood. This thesis presents three studies that explore the acoustic-phonetic adaptations of speakers in noisy and degraded communication conditions and their relationship with intelligibility. Study 1 investigated the effects of temporally fluctuating maskers on global acoustic-phonetic measures associated with speech in noise (Lombard speech). The results replicated findings of increased power in the modulation spectrum in Lombard speech, but showed little evidence of adaptation to masker fluctuations via the temporal envelope. Study 2 collected a larger corpus of semi-spontaneous communicative speech in noise and other degradations perturbing specific acoustic dimensions. Speakers showed different adaptations across the environments that were likely suited to overcome noise (steady and temporally fluctuating), restricted spectral and pitch information by a noise-excited vocoder, and a sensorineural hearing loss simulation. Analyses of inter-speaker variation in both studies 1 and 2 showed behaviour was highly variable and some strategy combinations were identified. Study 3 investigated the intelligibility of strategies ‘tailored’ to specific environments and the relationship between intelligibility and speaker acoustics, finding a benefit of tailored speech adaptations and discussing the potential roles of speaker flexibility, adaptation level, and intrinsic intelligibility. The overall results are discussed in relation to models of communication in adverse conditions and a model accounting for individual variability in these conditions is proposed