TL;DR
This study investigates how our speech mechanics change when we talk in noisy environments, a phenomenon known as the Lombard effect. Researchers found that speakers don't just get louder; they also adjust their lip movements, particularly increasing lip spreading and aperture, to compensate for the noise. These articulatory changes are strongly correlated with acoustic shifts, suggesting a coordinated effort to maintain intelligibility.
What Problem Does This Paper Address?
Communicating in noisy environments is challenging because our own voices are attenuated, and our speech becomes harder for listeners to understand. The Lombard effect describes the involuntary changes in speech production that occur in response to background noise. While previous research has extensively documented the acoustic changes (like increased loudness and pitch), there's been less focus on the specific articulatory adjustments that accompany these acoustic shifts. This paper aims to bridge that gap by simultaneously examining acoustic and articulatory changes in Lombard speech, asking: what specific lip movements are involved, how do they relate to acoustic changes, and do they differ based on the type of noise?
Methods
The researchers recorded a native French speaker producing short sentences in three conditions: silence, 85dB white noise, and 85dB cocktail party noise. They employed a multi-modal approach, capturing:
- Audio: Standard speech recordings.
- EGG (Electroglottography): To analyze vocal fold vibration and extract fundamental frequency (F0) and open quotient (Oq).
- Video: High-resolution video of the speaker's lips, analyzed using a labiometric device to extract key articulatory parameters.
The primary articulatory parameters analyzed were:
- Lip Spreading (A): How much the lips are pulled horizontally.
- Lip Aperture (B): The vertical opening of the mouth.
- Inter-lip Area (S): The area enclosed by the lips.
They also looked at "lip pinching" (lip compression when the mouth is closed) and the velocity peaks of these movements, which can indicate articulatory effort. Acoustic parameters like intensity, F0, and word duration were also extracted.
Datasets
The study used a custom-recorded corpus of 33 short French sentences with a subject-verb-object structure. The noisy environments were created using sounds from the BD Bruit database [11].
Key Insights and Ideas
- Hyper-articulation is real: Speakers don't just speak louder; they actively change their mouth movements.
- Lips are key players: The most significant articulatory changes involve increased lip aperture and spreading, not necessarily lip pinching.
- Articulatory effort matters: The speed of these lip movements (velocity peaks) also increases, suggesting greater physical effort.
- Noise type influences adaptation: Cocktail party noise elicits a greater increase in articulatory parameters than white noise, though white noise can lead to more pronounced spectral energy increases.
- Acoustic and articulatory changes are linked: Strong correlations were found between lip movements and acoustic features like F0 and spectral energy, indicating a coordinated system.
- Vowels get a boost: Acoustic adaptations in noise tend to favor vowels over consonants, potentially to help listeners reconstruct the message.
Key Results
- Lip Aperture (B) and Spreading (A) increased significantly in both noisy conditions, with a greater increase observed in cocktail party noise.
- Lip pinching decreased in white noise.
- Velocity peaks (Vmax) of lip movements (A, B, S) increased significantly in noise, with larger increases for B, S, and protruded lip pinching in cocktail party noise.
- Acoustic parameters confirmed previous findings: Voice intensity, F0, and word duration increased significantly in noise.
- Spectral energy increased significantly, particularly in the higher frequency bands (1500-3500Hz and 3500-5500Hz).
- Vowel intensity and duration increased more than consonant intensity and duration in noisy conditions.
- Strong correlations were observed between the maximum amplitudes of A and B (r = .91) and their velocity peaks (r = .70). F0 and spectral energy were also correlated with maximal amplitude of B.
Limitations and Future Work
- Single Speaker: The study was conducted with only one female native French speaker, limiting the generalizability of the findings. Future work should involve multiple speakers to explore individual differences and confirm these results.
- Specific Corpus: The corpus consisted of short SVO sentences. Further research with a more diverse range of sentence structures and speech tasks is needed.
- Vowel Specificity: While some vowel-specific analyses were performed, a more in-depth study on how different vowels are affected by Lombard speech is planned.
- Coarticulation: The authors acknowledge the need to explore hyper-articulation in relation to coarticulation more rigorously.
- Tongue Movements: The study focused on lip movements. Future work could investigate compensatory tongue movements, especially for rounded vowels with increased lip aperture.
- Generalizability: The authors are currently recording a new corpus with multiple speakers to assess the generalizability of their findings.
Summary updated