TL;DR
This study investigates how prosody, specifically accent and phrase position, influences the acoustic cues that distinguish consonant voicing and place of articulation in American English radio news speech. The findings reveal that accent generally strengthens these cues, enhancing the voicing contrast but not consistently the place contrast, and that prosodic strengthening prioritizes phonological function over pragmatic function.
What Problem Does This Paper Address?
While it's well-established that prosody (like pitch, loudness, and duration) conveys information about sentence structure and emphasis, its impact on the fine-grained acoustic details of speech sounds themselves is less understood, especially in naturalistic speech. Previous research often relied on "lab speech" (highly controlled, decontextualized utterances), leaving a gap in our understanding of how prosody shapes articulation in real-world communication. This paper aims to bridge that gap by examining prosodic effects on stop consonant acoustics in radio news broadcasts, a context that balances naturalness with a degree of clarity and structure. Specifically, the authors want to know if prosodic strengthening, a phenomenon where speech gestures become more pronounced in prominent positions, enhances the acoustic distinctions that signal consonant voicing (e.g., /p/ vs. /b/) and place of articulation (e.g., /p/ vs. /t/ vs. /k/).
Methods
The researchers analyzed acoustic properties of stop consonants (/p, t, k, b, d, g/) from the Boston University Radio News corpus. They focused on four acoustic measures: Voice Onset Time (VOT), fundamental frequency (f0) at vowel onset, closure duration (CD), and burst amplitude (BA). They also analyzed burst spectrum measures to assess place of articulation. The core of their analysis involved statistical modeling (2-way ANOVA) to determine the effects of two primary prosodic factors: accent (a level of phrasal prominence) and prosodic position (specifically, phrase-initial vs. phrase-medial). They also considered the interaction between accent and voice, and accent and place of articulation. To understand whether prosodic effects enhance or simply amplify existing contrasts, they examined the "effect size" (partial eta-squared, η²) of different factors.
Datasets
The study utilized the "lab news" portion of the Boston University Radio News corpus. This corpus consists of radio news stories read by professional announcers, transcribed and prosodically labeled using the ToBI standard. The dataset included approximately 42 minutes of speech from four announcers, sampled at 16kHz.
Key Insights and Ideas
- Prosody as a "Clarity Enhancer": The study strongly suggests that prosody doesn't just add emphasis; it actively shapes the acoustic realization of speech sounds to make them clearer, particularly for phonological contrasts.
- Voice Contrast Enhancement: Prosodic strengthening under accent demonstrably enhances the acoustic distinction between voiced and voiceless stops, making the voicing contrast more robust.
- Place Contrast - A Mixed Bag: While some acoustic cues for place of articulation are affected by accent, the overall enhancement of the place contrast is less consistent than for voicing. This suggests that the acoustic cues for place might be less susceptible to prosodic modulation for contrast enhancement.
- Phonological Function Trumps Pragmatic Function: A key finding is that prosodic strengthening effects are strongest on acoustic measures that are weaker cues for a given phonological contrast (e.g., CD for voicing, burst spectrum for place). This implies that the primary phonological role of an acoustic cue takes precedence over its secondary pragmatic role when prosody is applied.
- Phrase-Initial Position - Precision, Not Necessarily Contrast: While phrase-initial position didn't show the same pattern of contrast enhancement as accent, it did correlate with reduced acoustic variability, suggesting a different kind of prosodic strengthening focused on precision rather than amplified distinction.
Key Results
- Accent's Impact: Accent significantly affected VOT, f0, and CD for most speakers, with effects on BA and burst spectrum measures also observed.
- Voice Cues: Voice was a strong predictor of VOT and f0 variation across all speakers. Accent also significantly influenced these cues, often leading to greater separation between voiced and voiceless stops.
- Place Cues: POA was significantly encoded by VOT, CD, BA, and burst spectrum measures. Burst spectrum measures, particularly BS-hi, were most strongly affected by POA. Accent's effect on POA cues was more uniform, leading to less consistent contrast enhancement.
- Contrast Preservation: Accent effects were generally contrast-preserving, meaning they didn't obliterate existing distinctions. They tended to enhance the voicing contrast and showed uniform strengthening for place cues.
- IP-Initial Position: No significant effect of phrase-initial position was found on acoustic voicing cues for /t/ and /d/. However, IP-initial consonants showed reduced acoustic variability, suggesting greater articulatory precision.
Limitations and Future Work
- Data Limitations: The study acknowledges that the number of tokens for stops in phrase-initial position was too low for a complete analysis of positional effects. The uneven distribution of stops across prosodic conditions and the lack of control for adjacent vowel contexts are also noted.
- Function Words: While the inclusion of function words didn't substantively alter the main effects, their presence might introduce some noise.
- Future Research: The authors suggest further research is needed on the role of CD and BA as voice cues, as their findings were somewhat inconsistent with prior literature. They also highlight the need for perception studies to directly link acoustic distinctiveness to perceptual clarity and to determine the relative contribution of various acoustic cues. Investigating the role of formant transitions in POA encoding and exploring prosodic effects on vowel duration and intensity would also be valuable.
Summary updated