Intelligibility of speech spoken in noise/reverberation for older adults in reverberant environments

Nao Hodoshima, Takayuki Arai, Kiyohiro KurisuInterspeech 2012

TL;DR

This paper investigates how to make public space announcements more intelligible for older adults, who struggle more in noisy and reverberant environments. They found that speech produced with a "Lombard-like" effect (speech modified by noise/reverberation) significantly improves intelligibility for older listeners, even when the recording and listening conditions differ. This suggests a practical way to enhance public announcements without costly acoustic treatments.

What Problem Does This Paper Address?

As the global population ages, ensuring that critical information, such as public announcements in places like train stations and airports, is accessible to older adults becomes increasingly important. Older adults experience a significant decline in speech intelligibility, particularly in environments with background noise and reverberation. This paper tackles the challenge of improving the intelligibility of speech in such challenging acoustic conditions, specifically for older adults. The authors draw upon the "Lombard effect," a natural phenomenon where speakers unconsciously modify their speech to be more robust in noisy environments. They explore whether this modified speech, when used in public address systems, can enhance intelligibility for older listeners, even when the acoustic conditions during recording differ from the listening environment.

Methods

The study employed a listening test with 24 older adults. Speech materials consisted of target words embedded in carrier sentences, recorded under three speaking conditions:

  • Quiet (Q): Standard speech.
  • Noise (N): Speech produced while the speaker heard white noise through headphones (Noise-Induced Speech - NIS).
  • Reverberation (R): Speech produced while the speaker heard reverberant speech through headphones (Reverberation-Induced Speech - RIS).

These recorded speech materials were then convolved with two different impulse responses (R1 and R2) to simulate public spaces with reverberation times (RT) of 1.4 s and 2.4 s, respectively. Participants listened to these stimuli and were asked to write down the target words. The intelligibility was measured as the percentage of correctly identified morae. Statistical analysis (ANOVA) was used to determine the effects of speaking condition, reverberation time, and speaker on intelligibility. A key aspect of the methodology was testing the robustness of NIS/RIS by allowing differences between the speaking (recording) and listening reverberant conditions.

Datasets

The speech materials were based on a database of familiarity-controlled Japanese word lists (FW03) [13]. The specific target words were four-mora units selected for their familiarity. The impulse responses (R1 and R2) used to simulate reverberant environments were created by modifying exponential decays of an existing impulse response, aiming to mimic public spaces like subway stations or airports. No new datasets were introduced; existing resources were adapted for the experimental setup.

Key Insights and Ideas

  • Lombard Effect for Older Adults: The study provides strong evidence that speech modified by noise or reverberation (NIS/RIS) is significantly more intelligible for older adults in reverberant environments compared to standard speech. This suggests that leveraging the Lombard effect is a viable strategy for improving public announcements.
  • Robustness to Acoustic Mismatch: A crucial finding is that NIS/RIS remains more intelligible even when the acoustic conditions during recording (e.g., presence of noise or reverberation) differ from the listening environment. This implies that announcements don't need to be recorded under the exact same acoustic conditions as the public space they will be broadcast in.
  • Speaker Variability Matters: The effectiveness of NIS/RIS varied between the two speakers tested. This highlights the importance of individual speaker characteristics and suggests that identifying "intelligible speakers" or their acoustic traits could be beneficial for future applications.
  • Older Adults Benefit More: While both young and older adults benefit from RIS, the improvement in intelligibility for older adults was notably higher than for young adults in a previous study [11]. This suggests that older adults may gain a greater advantage from these speech modification techniques.

Key Results

  • Overall Intelligibility Improvement: For older adults, RIS was significantly more intelligible than speech spoken in quiet (Q) across both reverberant conditions (RT=1.4s and 2.4s). NIS was also more intelligible than Q for one speaker (S1) in the longer reverberation condition (RT=2.4s).
  • Effect of Reverberation Time: Intelligibility generally decreased as reverberation time increased, a significant main effect across all conditions.
  • Speaker Differences: Speaker S2 generally showed higher intelligibility than S1, and the effect of RIS was more pronounced for S2.
  • Acoustic Modifications: RIS for speaker S2 showed a significant increase in pitch and a trend towards increased F1 and consonant duration compared to quiet speech, aligning with known Lombard effect characteristics.

Limitations and Future Work

  • Limited Number of Speakers: The study used only two speakers, making it difficult to definitively attribute observed differences to speaker characteristics versus listener or environmental factors. Future work should involve a larger number of speakers to systematically investigate these relationships.
  • Specific Acoustic Conditions: The study focused on specific types of noise and reverberation. Further research could explore a wider range of acoustic conditions and their impact on intelligibility.
  • Identifying "Intelligible Speakers": The authors suggest that identifying speakers who naturally produce more intelligible NIS/RIS or characterizing the acoustic features that contribute to this intelligibility would be valuable for developing better speech synthesis or recording strategies.
  • Application to Speech Synthesis: While the study focused on recorded speech, future work could explore applying these principles to speech synthesis systems to create announcements that are inherently more intelligible for older adults in challenging environments.

Summary updated

BibTeX
@inproceedings{hodoshima2012intelligibility,
  author = {Nao Hodoshima and Takayuki Arai and Kiyohiro Kurisu},
  title = {{Intelligibility of speech spoken in noise/reverberation for older adults in reverberant environments}},
  booktitle = {Interspeech},
  year = {2012},
}
Figure