TL;DR
This paper investigates how the distinction between voiced and voiceless consonants is maintained in whispered speech, a mode of communication that lacks vocal fold vibration. The researchers found that while vocal fold vibration is absent, speakers still utilize subtle aerodynamic and durational cues, particularly in consonant closure and preceding vowel length, to signal this phonological contrast, suggesting a complex interplay of production constraints and perceptual needs.
What Problem Does This Paper Address?
In normal, "modal" speech, the difference between voiced and voiceless consonants (like /b/ vs. /p/, or /d/ vs. /t/) is primarily signaled by vocal fold vibration. However, whispered speech, often used for discretion or in noisy environments, intentionally eliminates this vibration. This raises a crucial question: how do speakers and listeners manage to distinguish between voiced and voiceless consonants in whisper? Previous research has suggested that other phonetic cues, such as the duration of consonants and preceding vowels, and differences in intraoral pressure, might play a role. This paper aims to systematically investigate these cues in French whispered speech, specifically examining consonant durations, vowel durations, and intraoral pressure, and to test the "Glottal Resistance Hypothesis" as a potential explanation for observed aerodynamic differences.
Methods
The study employed two main experiments.
Experiment 1: Voicing-Dependent Durations This experiment analyzed the acoustic duration of consonants and their preceding vowels in both modal and whispered speech.
- Corpus: Recordings were made of 4 French speakers reading randomized lists of non-sense and real words. The words were designed to contain specific voiced/unvoiced obstruent pairs (e.g., /b-p/, /t-d/, /k-g/, /f-v/, /s-z/, /ʃ-ʒ/) in intervocalic positions.
- Measurements: Acoustic durations of consonants (specifically closure/steady-state and release phases) and preceding vowels were measured using Praat.
- Analysis: Repeated-measure ANOVAs were used to analyze the effects of phonation mode (modal vs. whisper), voicing (voiced vs. unvoiced), and manner of articulation (stop vs. fricative) on these durations.
Experiment 2: Glottal Resistance Hypothesis This experiment focused on intraoral pressure (Po) as an indirect measure of glottal adjustments, testing the idea that glottal resistance might differ for voiced and unvoiced consonants even in whisper.
- Corpus: Recordings were made of 8 French speakers producing labial obstruents (/p-b/, /f-v/) in various positions (initial, final, medial) within monosyllabic and bisyllabic words, in both modal and whispered speech.
- Measurements: Intraoral pressure (Po) was measured using a fine tube inserted into the mouth, capturing the peak pressure during consonant production. Oral airflow was also recorded.
- Analysis: Repeated-measure ANOVAs were used to analyze the effects of phonation mode, voicing, and syllable position on peak Po.
Datasets
The study utilized newly recorded speech data from French speakers. No pre-existing public datasets were used for training or evaluation. The corpus consisted of:
- Experiment 1: 4 speakers, producing ~12 target non-sense words and 36 filler words, repeated 5 times for each phonation mode.
- Experiment 2: 8 speakers, producing 72 isolated words (with specific labial consonant contexts), repeated once for each phonation mode.
Key Insights and Ideas
- Durational Cues Persist: Even without vocal fold vibration, the duration of consonants and preceding vowels still carries information about phonological voicing in whispered speech. This suggests that speakers actively maintain these durational contrasts.
- Consonant Closure is Key: For obstruents, the steady-state or closure phase (rather than the release phase) appears to be the primary durational cue for voicing in French whisper. This is a notable difference from modal speech where both phases can contribute.
- Intraoral Pressure Reveals Complex Dynamics: The intraoral pressure (Po) patterns in whisper are not simply a reduced version of modal speech. While Po is higher for unvoiced consonants than voiced ones in both modes (supporting the Glottal Resistance Hypothesis for voiced consonants), the change in Po from modal to whisper is different for voiced and unvoiced consonants.
- Conflicting Constraints: The observed Po patterns in whisper suggest a compromise between different needs: maintaining the voicing contrast (potentially requiring specific glottal adjustments) and producing a low-intensity, discreet speech signal. This leads to an increase in Po for voiced consonants and a decrease for unvoiced ones, which is not fully explained by the simple Glottal Resistance Hypothesis alone.
Key Results
- Consonant Duration: Unvoiced obstruents were significantly longer than voiced ones in both modal (118 ms vs. 78 ms) and whispered speech (136 ms vs. 99 ms). The difference in duration between unvoiced and voiced fricatives reduced slightly in whisper, while the difference for stops remained similar.
- Vowel Duration: Pre-consonantal vowels were significantly longer before voiced obstruents than before unvoiced ones in both modal and whispered speech.
- Intraoral Pressure (Po):
- In both modal and whispered speech, unvoiced labial obstruents had a higher peak Po than voiced ones (modal: 5.42 hPa vs. 3.56 hPa; whisper: 4.91 hPa vs. 4.02 hPa).
- The difference in Po between unvoiced and voiced consonants was significantly halved from modal (1.86 hPa) to whispered speech (0.89 hPa).
- Voiced consonants showed an increase in Po from modal to whispered speech (3.47 to 4.04 hPa), while unvoiced consonants showed a decrease (5.42 to 4.91 hPa).
Limitations and Future Work
- Limited Scope: The study focused on French and specific consonant types (obstruents). Further research could explore other languages and phonetic contrasts.
- Indirect Measures: Intraoral pressure is an indirect measure of glottal behavior. Direct optical measurements of the glottis during whispered speech would provide more definitive evidence.
- Perceptual Studies: While the study focused on production, future work could investigate how these production differences are perceived by listeners in whispered speech.
- Explaining Unvoiced Consonant Po: The decrease in Po for unvoiced consonants in whisper remains an open question, potentially related to perceptual needs for low-intensity speech or subtle glottal adjustments not fully captured by the Glottal Resistance Hypothesis. The authors suggest that these patterns might result from conflicting perceptual constraints related to intelligibility and discretion.
Summary updated