Visual Hearing Aids: Artificial Visual Speech Stimuli for Audiovisual Speech Perception in Noise

Abstract

Speech perception is optimal in quiet environments, but noise can impair comprehension and increase errors. In these situations, lip reading can help, but it is not always possible, such as during an audio call or when wearing a face mask. One approach to improve speech perception in these situations is to use an artificial visual lip reading aid. In this paper, we present a user study (N = 17) in which we compared three levels of audio stimuli visualizations and two levels of modulating the appearance of the visualization based on the speech signal, and we compared them against two control conditions: an audio-only condition, and a real human speaking. We measured participants' speech reception thresholds (SRTs) to understand the effects of these visualizations on speech perception in noise. These thresholds indicate the decibel levels of the speech signal that are necessary for a listener to receive the speech correctly 50% of the time. Additionally, we measured the usability of the approaches and the user experience. We found that the different artificial visualizations improved participants' speech reception compared to the audio-only baseline condition, but they were significantly poorer than the real human condition. This suggests that different visualizations can improve speech perception when the speaker's face is not available. However, we also discuss limitations of current plug-and-play lip sync software and abstract representations of the speaker in the context of speech perception.