Prosodic Alterations to Speech Produced in Noise

R Patel, P Nunez, J Whang


This pilot study sought to better understand the extent to which the Lombard effect is enhanced for semantically salient words within an utterance. Four monolingual, English speakers produced a set of high and low predictability (HP and LP) sentences and nonsense sentences (SSI) in two noise (60 dB and 90 dB SPL) conditions and in quiet. Their productions were analyzed acoustically to identify changes in duration, intensity and fundamental frequency (F0) on semantically salient and non-salient words. Preliminary results indicated that salient words were louder and higher in pitch. These differences were further enhanced in the loud condition (90 dB) for SSI and HP sentences, possible evidence of a linguistically biased Lombard effect. Communicating in noisy situations may require a graded change in acoustic cues that are influenced by the semantic content of the utterance. While considerably more data and evidence are required, these findings may shed light on segmental and suprasegmental interrelations and the information content of prosodic cues. This work has implications for the development of speech synthesis algorithms that incorporate models of speech in noise. Synthesizers capable of altering speech quality would enable AAC users to communicate effectively in everyday noise situations.