TL;DR
This paper investigates how competing, simplified speech-like sounds (vowel-analogs) interfere with understanding a target speech sound, particularly focusing on the transitions within vowels. They found that these distracters primarily cause "informational masking" (making it harder to discern the target's features) rather than "energetic masking" (simply drowning out the sound), and that older adults require a significantly higher signal-to-noise ratio to achieve the same performance as younger adults.
What Problem Does This Paper Address?
Understanding speech in noisy environments, often referred to as the "cocktail party effect" (CPE), is a fundamental challenge for human auditory perception. While young, healthy individuals can often segregate a target speaker from background noise even at unfavorable signal-to-noise ratios (SNRs), this ability significantly deteriorates with age, even in individuals with otherwise normal hearing. This age-related decline in speech understanding in noise is thought to stem from deficiencies in auditory processing, particularly in temporal aspects like modulation masking and formant transition discrimination.
To better understand these age-related deficits and the mechanisms of auditory scene analysis (ASA), researchers often use simplified "speech-analog" stimuli. These stimuli retain key acoustic features of speech, like formant transitions, without the full complexity of natural speech. This paper specifically addresses how a "distracter" sound, also mimicking vowel transitions but with a different fundamental frequency, interferes with the perception of a target vowel transition. The authors aim to differentiate between "informational masking" (where the distracter interferes with the information in the target) and "energetic masking" (where the distracter simply raises the audibility threshold of the target). They also investigate how these masking effects differ between young and elderly listeners.
Methods
The researchers created synthetic, single-formant harmonic complex sounds that mimicked vowel transitions. These sounds consisted of two streams: a target and a distracter.
- Stimuli: Both target and distracter streams had dynamically changing formant frequencies. The target stream had a duration of 400 ms, while the distracter was 800 ms and started 200 ms earlier.
- Fundamental Frequency (f0): The two streams had different fundamental frequencies (107 Hz vs. 136 Hz or 189 Hz). Two frequency separations were used: a large one (Δf0/f0 = 0.77) and a narrow one (Δf0/f0 = 0.27).
- Formant Trajectories: The formant frequency in the target stream was modulated linearly (on an ERB scale) for 100 ms in one direction and then 100 ms in the opposite direction, creating an "up-down" or "down-up" pattern. The distracter stream had a constantly modulated formant frequency using a triangular waveform with the same modulation frequency and center frequency as the target.
- Masking Conditions:
- Informational Masking: Subjects were asked to identify the direction of the target formant trajectory (ascending then descending, or vice versa) while ignoring the distracter. The distracter level was varied, and the target level was adaptively adjusted to find the threshold for correct discrimination.
- Energetic Masking (Control): In a separate experiment, the threshold for detecting the target was measured, which is more indicative of energetic masking.
- Participants: Five young (average age 22.0 ± 3.4 years) and eight elderly (average age 69.0 ± 6.7 years) participants were tested. All participants had hearing thresholds within normal limits for the frequencies relevant to the formant changes.
Datasets
Not discussed in the paper. The stimuli were synthetically generated.
Key Insights and Ideas
- Informational Masking Dominates: The primary interference observed was informational masking. This means the distracter didn't just make the target harder to hear; it actively confused the listener about the target's specific acoustic features.
- Transition Shape Matters: The degree of masking was strongly influenced by the extent of the formant frequency excursion (the "swing"). Larger, more phonemically significant transitions were less susceptible to masking than smaller, sub-phonemic changes.
- Age-Related Deficits are Significant: Elderly listeners consistently required a higher target-to-distracter ratio (about 10-20 dB higher) to achieve the same performance as younger listeners, highlighting a substantial age-related decline in auditory scene analysis.
- Frequency Separation Plays a Role: While a larger frequency separation between the target and distracter generally led to better performance, this benefit was somewhat mitigated by the distracter's spectral overlap, especially in the narrow separation conditions.
- Energetic Masking is Less Relevant Here: The study confirmed that for these types of stimuli, informational masking was a much stronger effect than energetic masking, especially when the formant transitions were small.
Key Results
- Masking Increases with Distracter Level: As the distracter sound became louder, the threshold for discriminating the target's formant transition increased, indicating more masking. This effect was generally nonlinear, with smaller increases in target threshold for larger increases in distracter level, particularly for younger listeners and larger formant swings.
- Formant Swing Impact: Performance improved as the formant frequency excursion increased. This improvement was more rapid for smaller swings and plateaued for larger swings.
- Age Difference: Elderly listeners required approximately 10-20 dB higher target levels (i.e., a better SNR) compared to young listeners across most conditions.
- Informational vs. Energetic Masking: In conditions with large formant frequency excursions, the difference between informational and energetic masking was negligible. However, with small formant excursions, informational masking exceeded energetic masking by 20 dB or more.
Limitations and Future Work
- Stimulus Simplification: The use of single-formant vowel analogs, while useful for isolating specific acoustic features, is a simplification of natural speech. Future work could explore these effects with more complex, multi-formant stimuli.
- Sub-phonemic Transitions: The paper highlights that masking of small, sub-phonemic formant transitions could lead to errors in identifying consonants that precede or follow vowels. This suggests that while listeners might still identify the vowel identity, their overall speech intelligibility could be compromised. Future research could directly investigate the impact of this masking on consonant identification and overall sentence intelligibility.
- SNR Requirements for Elderly Listeners: The finding that elderly listeners require significantly higher SNRs underscores the need for more robust speech processing technologies and assistive listening devices tailored to their specific auditory challenges.
- Spatial Separation Ignored: This study focused on masking in a single spatial channel. Future work could explore how spatial separation, which benefits young listeners in CPE, interacts with these masking effects in older adults.
Summary updated