ISCA Archive Interspeech 2023
ISCA Archive Interspeech 2023

Audio, Visual and Audiovisual intelligibility of vowels produced in noise

Maeva Garnier

Why do speakers amplify articulatory movements when communicating in noisy environments? This study examines the hypothesis that hyper-articulation contributes to improved vowel intelligibility in audio, visual and audiovisual domains. A perceptual test was conducted with Audio-Only (AO), Visual-Only (VO) and Audiovisual (AV) stimuli of vowels produced in conversational and Lombard speech. The average score of vowel recognition was significantly increased in Lombard speech, compared to normal speech, for all perceptual modalities (AO, VO and AV). Specifically, the distinctive features of vowel height and backness were better perceived in Lombard speech in both the audio and visual domains. Changes in speech articulation in noise did not affect the perception of the rounding feature in the visual domain, but degraded it in the audio domain. On the contrary, the perception of the spreading feature was decreased in Lombard speech in the visual domain, but improved in the audio domain.