TL;DR
This study investigates how native Mandarin and Thai speakers perceive pitch contours in synthesized speech. Surprisingly, both groups struggled to distinguish falling from level pitch contours, but were much better at differentiating rising contours from either falling or level ones, suggesting a potential inherent salience of rising pitch.
What Problem Does This Paper Address?
Understanding how humans perceive pitch is crucial for fields ranging from speech synthesis to the study of tonal languages. While it's known that native speakers of tonal languages are generally better at distinguishing lexical tones than non-native speakers, the specific perceptual salience of different types of pitch contours (rising, falling, level) among native tone listeners themselves remains less explored. This paper aims to fill this gap by comparing the discrimination abilities of native Mandarin and Thai speakers on synthesized speech stimuli with these distinct pitch contours.
Methods
The researchers employed a same-different categorial discrimination task. Participants listened to three-stimulus trials (ABX) and had to determine if the pitch contour of the third stimulus (X) was the same or different from the first two (A and B). The stimuli were synthesized nonsense CV (/ba:/) syllables with linearly rising, falling, or level fundamental frequency (F0) contours. To ensure comparability, the F0 onset and offset difference for rising and falling contours was set at 10 Hz, and all stimuli were normalized for peak intensity. Perceptual sensitivity was quantified using d-prime scores, which were then analyzed using a repeated measures ANOVA.
Datasets
The study did not use pre-existing datasets for training or evaluation. Instead, they created their own dataset of synthesized speech stimuli. The stimuli consisted of 17 tokens each of linearly rising, falling, and level F0 contours, derived from a single male Assamese speaker's /ba:/ syllable, modified using the Praat software.
Key Insights and Ideas
- Rising Contour Salience: The most striking finding is the consistent and significantly better discrimination of rising pitch contours compared to falling or level contours, regardless of the listener's native language (Mandarin or Thai). This suggests that rising pitch might possess an inherent psychoacoustic salience.
- Falling-Level Confusion: Both Mandarin and Thai listeners showed a notable difficulty in distinguishing between falling and level pitch contours. This suggests that the subtle differences between these two might be less perceptually distinct for native tone listeners.
- Native Language Influence (Subtle): While the overall discrimination patterns were similar between Mandarin and Thai listeners, the paper hints that native language experience may play a role, though the study's design didn't yield statistically significant differences in this regard. The discussion points to how falling tones in both Thai and Mandarin have specific acoustic characteristics (e.g., greater F0 drop) that might influence perception.
Key Results
- A repeated measures ANOVA revealed a significant main effect of pitch contour contrast on d-prime scores, indicating that some contrasts were easier to discriminate than others.
- Post-hoc analysis showed that the d-prime scores for the falling-level (FL) contrast were significantly lower than for other contrasts (p < .05).
- Similarly, the level-falling (LF) contrast also yielded significantly lower d-prime scores than the rising-falling (RF) contrast (p < .05).
- While not statistically significant, the trend indicated that listeners were better at discriminating rising contours from both falling and level contours.
Limitations and Future Work
The authors acknowledge that the stimuli were synthesized and not natural speech, which might limit the generalizability of the findings. They also note that while native language experience is hypothesized to play a role, the study did not find statistically significant differences between Mandarin and Thai listeners. Future work is suggested to further confirm the hypothesis of rising contour salience through additional experiments. Investigating the influence of other acoustic parameters and the functional load of tones in different languages could also provide deeper insights.
Summary updated