Abstract
A set of experiments was performed to make a cross-language comparison of intelligibility of locally time-reversed speech, employing a total of 117 native listeners of English, German, Japanese, and Mandarin Chinese. The experiments enabled to examine whether the languages of three types of timing—stress-, syllable-, and mora-timed languages—exhibit different trends in intelligibility, depending on the duration of the segments that were temporally reversed. The results showed a strikingly similar trend across languages, especially when the time axis of segment duration was normalised with respect to the deviation of a talker’s speech rate from the average in each language. This similarity is somewhat surprising given the systematic differences in vocalic proportions characterising the languages studied which had been shown in previous research and were largely replicated with the present speech material. These findings suggest that a universal temporal window shorter than 20–40 ms plays a crucial role in perceiving locally time-reversed speech by working as a buffer in which temporal reorganisation can take place with regard to lexical and semantic processing.
Highlights
Speech reversal[1, 2] is a simple but effective technique in investigating speech perception or speech-related effects
When these variables were negligible, i.e., the original natural speech was produced at a moderate rate, and intelligibility was objectively measured with native listeners, the curves exhibited a rather similar trend; the 50%-intelligibility points fell in the range of 60–80 ms (Supplementary Fig. S2)
Separate analysis of variance (ANOVA) on arcsine-transformed intelligibility, specifying talkers and participants as random effects, revealed that the main effect of the segment duration was significant in all languages [American English: F(6, 946) = 864.52, p < 0.01; German: F(6, 946) = 613.29, p < 0.01; Japanese: F(6, 946) = 1170.59, p < 0.01; Mandarin Chinese: F(6, 465) = 361.16, p < 0.01]
Summary
Speech reversal[1, 2] is a simple but effective technique in investigating speech perception or speech-related effects It produces pairs of speech-based stimuli that are matched in their long-term spectra, but yield opposite results regarding intelligibility: Playing a recorded spoken sentence backwards makes it totally unintelligible[3]. Remez et al.[15] provided some evidence that subjective ratings could be a source of bias to make an intelligibility curve shallower than a curve obtained with a more objective method, along with presenting much deteriorated performance with locally time-reversed sine-wave speech, in which spectral information was greatly reduced These results suggest that familiarity with a specific language, speech rates, and experimental methods are possible sources of variability, with regard to the intelligibility of natural speech that is segmented and reversed in time. The inconsistencies may look subtle, they can be critical when considering brain mechanism
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
