Older adult’s identification and memory of synthetic and natural speech in noise

Abstract

This research tested older adult’s performance on identification and recall when presented sentences taken from the Speech Perception in Noise Test [Bilger et al., J. Speech Hear. Res. 27, 32–48 (1984)] synthesized using AT&T Natural Voices speech synthesizer. Performance on synthesized speech was compared to identification and recall of natural speech in noise. In experiment 1, participants were instructed to identify the final word of high- and low- predictability natural and synthesized sentences presented at three signal-to-noise ratios. The results showed that natural speech was easier to identify than synthesized speech at every speech-to-noise ratio for high and low predictability sentences. In addition, replicating Pichora-Fuller et al. [J. Acoust. Soc. Am. 97, 593–608 (1995)], pure tone thresholds were significantly correlated with identification performance for both natural and synthetic speech. In experiment 2, participants were given a sentence span task in which they were presented two to five natural and synthesized sentences and asked to recall the final words. These experiments show that, even with high context, synthetic speech is more difficult to understand and remember than natural speech. These results highlight concerns about the use of synthesized speech in assistive devices, particularly for elderly listeners. [Work supported by NIH.]