Skip to main content
. 2021 Oct 27;41(43):8946–8962. doi: 10.1523/JNEUROSCI.2891-20.2021

Figure 2.

Figure 2.

Speech feature correlation analysis for TIMIT and movie trailers. A, Stimulus correlations within TIMIT show the co-occurrence of features within the TIMIT stimuli. For example, sonorant phonemes are likely to be voiced and syllabic. The co-occurrence of specific phonological features with spectral and pitch features was low, since sentences were spoken by a variety of speakers. B, Stimulus correlations within the movie trailers. Co-occurrence of phonological features was highly similar to TIMIT. The spectrogram features were more correlated with one another, likely because of the presence of background noise, music, and other sounds in tandem with speech. C, The difference in stimulus correlation values between TIMIT and movie trailers. Overall, phonological feature correlations were very similar, but differences were observed in the co-occurrence of low- and high-spectrotemporal information, with TIMIT showing separate epochs with low- or high-frequency content (but not both), and movie trailers showing epochs with frequencies across the spectrum. D, Average stimulus correlations for acoustic and linguistic features in TIMIT and movie trailers. Acoustic features were generally more correlated than the phonological features, but the degree of correlation across stimulus sets was relatively similar.