This study investigates how prosodic prominence mediates the perception of American English vowels, testing the effects of F0 and duration. In Experiment 1, the perception of four vowel continua varying in duration and formants (high: /i-ɪ/, /u-ʊ/, non-high: /ɛ-ae/, /ʌ-ɑ/), was examined under changes in F0-based prominence. Experiment 2 tested if cue usage varies as the distributional informativity of duration as a cue to prominence is manipulated. Both experiments show that duration is a consistent vowel-intrinsic cue. F0-based prominence affected perception of vowels via compensation for peripheralization of prominent vowels in the vowel space. Longer duration and F0-based prominence further enhanced the perception of formant cues. The distributional manipulation in Experiment 2 exerted a minimal impact. Findings suggest that vowel perception is mediated by prominence in a height-dependent manner which reflects patterns in the speech production literature. Further, duration simultaneously serves as an intrinsic cue and serves a prominence-related function in enhancing perception of formant cues.
It has been well known that rising/falling pitch is employed to distinguish the rising (R) or falling (F) tones from the high-level (H) tone in Mandarin, but whether F0 range or F0 slope is the more critical F0 cue to perception is still inconclusive. To clarify this issue quantitatively, we took the F tone as the test case, and conducted two-alternative forced choice identification tests on two types of two-dimensional high-level–falling (H-F) tonal continua, one of which was manipulated along F0 range and duration (“F0 range continuum”) while the other along F0 slope and duration (“F0 slope continuum”). Experimental results indicated that F0 range was the primary cue because it resulted in a more robust (less duration-dependent) perceptual boundary than F0 slope. Meanwhile, the perceptual boundary in F0 range was not fully independent of but mildly modulated by duration, suggesting that duration (or equivalently, F0 slope) played a supplementary role in identifying the H-F tonal contrast.
The acoustic cues for prosodic prominence have been explored extensively, but one open question is to what extent they differ by context. This study investigates the extent to which vowel type affects how acoustic cues are related to prominence ratings provided in a corpus of spoken Mandarin. In the corpus, each syllable was rated as either prominent or non-prominent. We predicted prominence ratings using Bayesian mixed-effect regression models for each of four Mandarin vowels (/a, i, ɤ, u/), using fundamental frequency (F0), intensity, duration, the first and second formants, and tone type as predictors. We compared the role of each cue within and across the four models. We found that overall duration was the best predictor of prominence ratings and that formants were the weakest, but the role of each cue differed by vowel. We did not find credible evidence that F0 was relevant for /a/, or that intensity was relevant for /i/. We also found evidence that duration was more important for /ɤ/ than for /i/. The results suggest that vowel type credibly affects prominence ratings, which may reflect differences in the coordination of acoustic cues in prominence marking.
From a very brief speech, human listeners can estimate the pitch range of the speaker and normalize pitch perception. Spectral features which inherently involve both articulatory and phonatory characteristics were speculated to play roles in this process, but few were reported to directly correlate with speaker’s F0 range. To mimic this human auditory capability and validate the speculation, in a preliminary study we proposed an LSTM-based method to estimate speaker’s F0 range from a 300 ms-long speech input, which turned out to outperform the conventional method. By two more experiments, this study further improved the method and verified its validity in estimating the speaker-specific underlying F0 range. After incorporating a novel measurement of F0 range and a multi-task training approach, Experiment 1 showed that the refined model gave more accurate estimates than the initial model. Based on a Japanese-Chinese bilingual parallel speech corpus, Experiment 2 found that the F0 ranges estimated with the model from the Chinese speech and the model from the Japanese speech produced by the same set of speakers had no significant difference, whereas the conventional method showed significant difference. The results indicate that the proposed spectrum-based method captures the speaker-specific underlying F0 range which is independent of the linguistic content.