Perceiving listener-directed speech: effects of authenticity and lexical neighborhood density

Rebecca Scarborough, Georgia ZellouInterspeech 2012

TL;DR

This study investigates how listeners perceive speech produced in different "clear speech" conditions, specifically authentic listener-directed speech versus simulated clear speech. The key finding is that authentic listener-directed speech is perceived faster, and surprisingly, words from dense phonological neighborhoods are also identified faster in this authentic condition, suggesting that real-world communicative adjustments are perceptually beneficial.

What Problem Does This Paper Address?

When we communicate, we often adjust our speech to be clearer, especially when we anticipate listener difficulties (e.g., noise, hearing impairment). This "clear speech" is generally characterized by hyperarticulation (more distinct sounds) and altered coarticulation (how sounds influence each other). However, the acoustic properties of clear speech can vary depending on whether it's a simulated scenario (like being told to "speak clearly") or authentic speech directed at a real listener. Furthermore, linguistic factors like word frequency and "lexical neighborhood density" (how many similar-sounding words exist) also influence speech production, with dense neighborhoods often leading to more hyperarticulation and coarticulation.

This paper tackles a crucial question: do these production adjustments, particularly those made in authentic listener-directed speech, actually help listeners perceive speech more effectively? Specifically, it asks if the acoustic modifications in real listener-directed speech are more beneficial than those in simulated clear speech, and how these effects interact with lexical neighborhood density.

Methods

The researchers used a lexical decision task to assess speech perception. Participants listened to recorded speech tokens and had to quickly decide if they were real words or nonsense words. The speech tokens were drawn from a previous production study where a single speaker produced words under two conditions:

  1. "Hard of Hearing" condition: Simulated clear speech, where the speaker was instructed to "speak as if to someone hard of hearing."
  2. Real Listener-directed speech: Authentic speech produced while interacting with a real listener.

These tokens were further categorized by lexical neighborhood density (ND):

  • High ND: Words with many phonologically similar neighbors.
  • Low ND: Words with few phonologically similar neighbors.

The experiment involved 9 participants, and response times (RTs) from correct responses were analyzed using a linear mixed-effects model. The model examined the effects of speech condition (Real Listener vs. Hard of Hearing) and neighborhood density (Hi ND vs. Lo ND) on RTs.

Datasets

The speech tokens used in this study were derived from a previous production study [8]. The words and nonwords were selected from the CELEX lexical database [11] for their phonological properties and lexical frequencies. Neighborhood density was calculated based on phonemic similarity and edit distance, using data adjusted for a general American English dialect. No new datasets were introduced; rather, existing data from a prior production study was repurposed for a perception experiment.

Key Insights and Ideas

  • Authenticity Matters: Real listener-directed speech is perceived significantly faster than simulated clear speech. This suggests that the subtle, natural adjustments made when speaking to an actual person are more perceptually beneficial than instructed "clear speech."
  • Dense Neighborhoods Can Be Faster: Counterintuitively, words from dense phonological neighborhoods (which are typically harder to distinguish due to competition) were identified faster in the authentic listener-directed speech condition. This implies that the acoustic modifications in this natural speech style were so effective that they overcame the inherent difficulty of dense neighborhoods.
  • Nasal Coarticulation is Key: The paper highlights that while both speech styles involved hyperarticulation, real listener-directed speech had more nasal coarticulation than the "hard of hearing" condition. This difference in coarticulation, rather than hyperarticulation alone, is proposed as a critical factor for the perceptual advantage of authentic speech.
  • Lexical Effects are Context-Dependent: The typical finding that low-density words are easier to perceive was reversed in the authentic listener-directed speech condition. This demonstrates that the communicative context can significantly alter how lexical factors influence perception.

Key Results

  • Listeners responded significantly faster to speech from the Real Listener-directed condition compared to the "Hard of Hearing" condition (p < 0.0001).
  • There was a significant interaction between speech condition and neighborhood density.
  • In the "Hard of Hearing" condition, listeners responded faster to Lo ND words than to Hi ND words (t(97)=4.52, p<.001), as expected.
  • However, in the Real Listener-directed condition, listeners responded more quickly to Hi ND words than to Lo ND words (t(96)=6.62, p<.001).

Limitations and Future Work

  • Single Speaker: The study used speech from only one female native English speaker. This limits the generalizability of the findings to other speakers and dialects.
  • Simulated vs. Real: While the "Hard of Hearing" condition is a common simulation, it might not fully capture the range of simulated clear speech. The contrast is primarily between a specific simulation and authentic interaction.
  • Future Work: The authors suggest that future research should explore the perceptual consequences of clear speech more broadly, investigating whether the observed pattern of neighborhood-conditioned effects in production could be listener-directed. Further studies could also examine how different types of listener difficulties (beyond simulated hearing impairment) influence these perceptual outcomes.

Summary updated

BibTeX
@inproceedings{scarborough2012perceiving,
  author = {Rebecca Scarborough and Georgia Zellou},
  title = {{Perceiving listener-directed speech: effects of authenticity and lexical neighborhood density}},
  booktitle = {Interspeech},
  year = {2012},
}
Figure