TL;DR
This study reveals that our speech rate adjustments in response to delayed auditory feedback (DAF) are not uniform. Instead, they are strategically influenced by the prosodic structure of the utterance, with accented syllables, particularly those at the boundaries of prosodic phrases, showing a greater tendency to lengthen. This suggests a sophisticated interplay between motor control, auditory feedback, and linguistic structure.
What Problem Does This Paper Address?
We know that when speakers hear their own voice with a delay (Delayed Auditory Feedback or DAF), they tend to slow down their speech, often by lengthening syllables. This is a fascinating demonstration of the brain's ability to adapt speech motor control based on sensory feedback. However, it's been unclear whether this adaptation is a general slowing down or if it's more nuanced. Specifically, researchers wondered if the position of a syllable within the rhythmic and intonational structure of an utterance (its prosodic structure) plays a role in how much it's affected by DAF. For instance, are syllables that are naturally emphasized or at the end of a phrase treated differently? This paper aims to disentangle these effects by investigating how DAF interacts with French prosodic structure, particularly accent and final lengthening.
Methods
The researchers used a controlled experimental setup involving 10 native French speakers. They recorded speakers reading three distinct five-syllable French sentences, each designed to have different syllabic complexity (CV, CVC, CCV(C)). The core of the experiment involved applying varying levels of DAF (0ms, 60ms, 120ms) to the speakers' audio feedback in real-time.
To analyze the data, they meticulously labeled vowel and sentence boundaries in baseline recordings and then used Dynamic Time Warping (DTW) with cepstral representations to align these labels across all repetitions. Statistical analysis was performed using linear mixed-effects models in R, allowing them to account for individual speaker variability. The key metrics analyzed were vowel duration and the ratio of durations between accented and non-accented vowels within specific prosodic units (first accentual phrase and final accentual phrase).
Datasets
The study utilized recordings of 10 native French speakers reading three specific five-syllable sentences. These sentences were designed to vary in syllabic complexity:
- CV sentence: /vivjẽ vi lə vẽ/ (Vivien vit le vin)
- CVC sentence: /zaklin zɛr lə zur/ (Jacqueline gère le jour)
- CCV(C) sentence: /bradle briz lə braа/ (Bradley brise le bras)
No new benchmark datasets were introduced; the focus was on analyzing existing speech production under manipulated auditory feedback.
Key Insights and Ideas
- Prosody is King: The most significant finding is that DAF's effect on speech timing is not uniform. It's strongly modulated by the prosodic structure of the utterance.
- Accentuation Matters: Accented vowels, especially those at the right edge of accentual phrases (the basic rhythmic units in French), are lengthened more by DAF than their non-accented counterparts.
- Boundary Effects are Amplified: Vowels at the boundaries of larger prosodic units, like intonational phrases (IPs), appear even more sensitive to DAF. This suggests that the natural tendency for final lengthening at IP boundaries interacts with DAF to produce greater duration changes.
- Not Just Duration: The increased lengthening of accented vowels isn't simply because they are inherently longer. The status of the vowel within the prosodic hierarchy seems to be the more critical factor.
- Melodic Contour's Role: While the melodic contour at the end of an intonational phrase (rising vs. falling) might influence final lengthening, its interaction with DAF's effect on accented vowels was found to be similar across contours.
Key Results
- General Lengthening: DAF generally leads to longer syllable durations, with the effect increasing with higher DAF levels (60ms and 120ms).
- Accentual Phrase Boundary: The ratio of accented to non-accented vowel duration (V2/V1) in the first accentual phrase increased with DAF, indicating greater lengthening of the accented vowel (V2).
- Intonational Phrase Boundary: The ratio of accented to non-accented vowel duration (V5/V4) in the final accentual phrase also showed increased lengthening, particularly with a 60ms DAF, suggesting heightened sensitivity at IP boundaries.
- Saturation Effect: At the highest DAF level (120ms), the lengthening effect on the final IP boundary vowels (V5/V4) seemed to plateau or even decrease, possibly due to a saturation of the natural final lengthening mechanism.
- Syllable Complexity Irrelevant: The degree of lengthening was not directly correlated with the inherent syllabic complexity (CV, CVC, CCV(C)) of the sentences, further supporting the idea that prosodic structure, not just segment duration, is the key modulator.
Limitations and Future Work
- Sentence Material: The study used relatively short, five-syllable sentences. The authors suggest that future work should employ longer sentence material with more intermediate accentual phrases and intonational phrases to better explore the complex interactions.
- Population Size: While 10 speakers were used, a larger and more diverse population would strengthen the generalizability of the findings.
- Language Comparison: The study focused on French. Comparing these results with languages that have different rhythmic properties (e.g., English) could provide broader insights into the universality of these prosodic-timing interactions.
- Segmental Factors: The analysis did not account for intrinsic vowel properties (like vowel height) that can influence duration and might interact with DAF effects.
- Melodic Contour Nuances: While the study explored rising vs. falling contours, further investigation into the precise role of melodic contours at IP boundaries and their interaction with DAF is warranted.
Summary updated