Effect of noise type and level on focus related fundamental frequency changes

Martti Vainio, Daniel Aalto, Antti Suni, Anja Arnhold, Tuomo Raitio, Henri Seijo, Juhani Järvikivi, Paavo AlkuInterspeech 2012

TL;DR

This study investigates how different types and levels of background noise affect the fundamental frequency (f0) contours of speech, particularly when speakers are marking specific words for emphasis (focus). The findings reveal that speakers adjust their pitch not only based on overall loudness but also on the specific characteristics of the noise, suggesting a more nuanced role for pitch in the Lombard effect than previously understood.

What Problem Does This Paper Address?

When speaking in noisy environments, people tend to speak louder, at a higher pitch, and with longer speech segments – a phenomenon known as the Lombard effect. This adaptation is crucial for maintaining intelligibility and ensuring effective communication. While previous research has explored how different noise levels and types influence the Lombard effect, there's a gap in understanding how these factors interact with linguistic features like "focus marking." Focus marking is how speakers highlight specific words or phrases to convey new or important information, often signaled by changes in pitch. This paper aims to disentangle the effects of noise type and level on these linguistically driven pitch changes, even when the perceived loudness of the noises is equated.

Methods

The researchers designed an experiment where 21 participants spoke short, three-word Finnish sentences in response to questions designed to elicit different focus conditions: broad focus, narrow focus on the first word (subject), and narrow focus on the last word (object). These utterances were recorded in an anechoic chamber while participants listened to three types of noise through closed headphones: babble noise (simulating informational masking), white noise (simulating energetic masking), and a 1 kHz low-pass noise (designed to differentially mask vowels). Crucially, these noises were carefully calibrated to have equal perceived loudness across three distinct levels (approximately 60, 70, and 80 dB(A)).

The core of the analysis focused on the fundamental frequency (f0) contours of the speech. They analyzed f0 at specific points within each word (peak and valleys) and calculated two key measures: the mean f0 level and the "Bounded Variation" (BV) norm, which quantifies the overall movement or expansion of the f0 contour throughout an utterance. Statistical analysis was performed using linear mixed-effects models to account for individual speaker and item variations, examining the effects of focus type, noise level, noise type, and gender.

Datasets

Not discussed in the paper. The study used pre-recorded noise signals from the NOISEX-92 database for the babble noise.

Key Insights and Ideas

  • Noise Type Matters Beyond Loudness: Even when noises are perceived as equally loud, their spectral characteristics significantly influence how speakers adjust their pitch. This suggests that the Lombard effect isn't solely driven by a general need to overcome loudness but also by how the noise interferes with specific acoustic cues, including pitch.
  • Pitch Plays a Special Role in Lombard Speech: The findings indicate that pitch (f0) changes are not just a byproduct of increased vocal intensity but are actively modulated by noise type and linguistic focus. This highlights pitch as a potentially more direct and adaptable signal in noisy communication.
  • Focus Marking Persists, but is Modified: While the general patterns of focus marking in Finnish (e.g., higher pitch on focused words) were observed, these patterns were subtly altered by the noise conditions. This suggests that speakers attempt to preserve communicative intent even under challenging acoustic circumstances.
  • Differential Masking Effects: The low-pass noise had a less pronounced effect on mean f0 compared to babble noise, while white noise led to greater f0 expansion. This aligns with the idea that different noise types mask different aspects of the speech signal, leading to varied compensatory strategies.

Key Results

  • Mean f0: Increased significantly with noise level. Different noise types also had a significant effect, with low-pass noise leading to lower mean f0 than babble noise. Male speakers had significantly lower mean f0 than female speakers.
  • f0 Expansion (BV norm): Also increased significantly with noise level, showing an exponential increase. White noise resulted in greater f0 expansion than babble or low-pass noise.
  • Focus Effects: Both mean f0 and f0 expansion were significantly influenced by focus type, with narrow focus on the first word (N1) generally leading to lower f0 and narrow focus on the last word (N2) leading to higher f0.
  • Interactions: Significant interactions were found, such as between noise type and noise level, indicating that the effect of noise level on f0 could depend on the type of noise.

Limitations and Future Work

  • Limited Linguistic Material: The study used very short, simple sentences. Future work could explore these effects with more complex linguistic structures and naturalistic speech.
  • Artificial Environment: Recordings were made in an anechoic chamber with headphones. While controlled, this doesn't fully replicate real-world listening conditions.
  • Focus on f0: While the paper focuses on f0, other prosodic features (intensity, duration) were also analyzed but not detailed in this excerpt. A more comprehensive analysis of all prosodic adjustments would be beneficial.
  • Future Directions: The authors suggest that the differential effects of noise types on pitch point to a specific importance of pitch in auditory feedback and that the Lombard effect might be as much about pitch adjustment as it is about intensity increase. Further research could explore the perceptual consequences of these noise-dependent pitch variations.

Summary updated

BibTeX
@inproceedings{vainio2012effect,
  author = {Martti Vainio and Daniel Aalto and Antti Suni and Anja Arnhold and Tuomo Raitio and Henri Seijo and Juhani Järvikivi and Paavo Alku},
  title = {{Effect of noise type and level on focus related fundamental frequency changes}},
  booktitle = {Interspeech},
  year = {2012},
}
Figure