Cognitive and Linguistic Interference in Speech Perception

127 concepts47 papers1 proposal

When you try to follow a single conversation at a loud cocktail party, your auditory system faces a daunting challenge. For decades, scientists modeled this challenge primarily as a physical problem: sound waves from different sources collide in the air and mix together in your inner ear, washing out the acoustic details of the target speaker.

However, a growing body of research demonstrates that the struggle to hear in noise is not just a peripheral physical limitation, but a deeply cognitive one. To understand how we navigate these environments, researchers divide auditory interference into two distinct mechanisms: energetic masking and informational masking.

  • Energetic Masking (EM) occurs at the physical level of the cochlea. It happens when the acoustic energy of a background noise overlaps with the frequencies of a target speech signal, effectively drowning it out. Think of trying to talk over a roaring waterfall or a roaring train engine.
  • Informational Masking (IM) occurs at higher levels of cognitive processing. It happens when the background noise is acoustically similar to speech or contains meaningful linguistic information that competes for the brain's attention and language-processing resources. Think of trying to read a book while someone next to you is having a juicy conversation in your native language.

Building an intuition for why informational masking works—and why it is so disruptive—requires looking at how the brain automatically processes human speech, how it gets confused by competing talkers, and why traditional engineering models struggle to predict it.


The Paradox of the Competing Talker

Intuitively, you might think that the more people talking in the background, the harder it is to understand a target speaker. But the relationship between the number of background talkers and speech intelligibility is non-linear and reveals the fundamental difference between EM and IM.

As Xihong Wu et al. demonstrated in Effect of number of masking talkers on speech-on-speech masking in Chinese, the steepest drop in a listener's ability to comprehend target speech occurs when moving from one to two competing background talkers.

Why does this happen?

When a hundred people talk at once (multi-talker babble), their individual voices blend into a continuous, stationary blanket of sound. This blanket acts almost entirely as an energetic masker. The brain doesn't try to parse the words of a hundred people simultaneously; it treats the crowd like a mechanical hum.

But when only one or two people are talking in the background, the noise is sparse and highly dynamic. As Fei Chen et al. explain in Effect of linguistic masker on the intelligibility of Mandarin sentences, sparse background speech leaves open acoustic "windows" through which fragments of words can be clearly heard. However, because these fragments sound like intelligible human speech, they act as a powerful informational masker. The brain's executive control is tricked into tracking these background streams, dramatically increasing cognitive load.


Obligatory Semantic Access and Linguistic Overlap

The core reason informational masking is so potent lies in a psychological phenomenon known as obligatory semantic access: when your brain hears words spoken in a language you understand, it cannot help but try to decode their meaning.

In a study simulating high-speed train environments, Weigang Wei et al. evaluated how passenger chatter versus mechanical train rumble affected concentration in Interference of Speech and Interior Noise of Chinese High-Speed Trains with Task Performance. They found that while listeners easily habituated to high-intensity stationary mechanical noise, irrelevant background speech significantly degraded performance on complex tasks. The volume of the noise mattered far less than its intelligibility.

To isolate exactly how much the "meaningfulness" of background noise hurts us, Mingyue Huo et al. conducted an experiment in Quantifying Informational Masking due to Masker Intelligibility in Same-talker Speech-in-speech Perception. By systematically degrading the clarity of a competing background voice while keeping its physical sound energy identical, the authors found that making a background voice clear enough to understand can slash a listener's target speech recognition by up to 50%.

This cognitive bottleneck becomes even tighter when the background noise shares linguistic properties with the target speech:

  1. Lyrical Interference: In The Conversation Continues: the Effect of Lyrics and Music Complexity of Background Music on Spoken-Word Recognition, Odette Scharenborg et al. showed that background music with sung lyrics impairs spoken-word recognition far more than complex instrumental music. The brain's linguistic processing hardware is forced to split its finite resources between the background lyrics and the foreground speaker.
  2. Language Similarity: The closer the background language is to the target language, the worse the interference. Aurore Gautreau et al. demonstrated in Let's all speak together! exploring the impact of various languages on the comprehension of speech in multi-linguistic babble that speech comprehension drops severely when the background babble is in the same language as the target. This occurs because the target and masker streams compete within the exact same phonetic, lexical, and semantic pathways.

Why Traditional Auditory Models Fail

For decades, audio engineers built objective metrics to predict whether a human would be able to understand speech in a given environment. Many of these models rely on the concept of auditory glimpsing.

If you look at a spectrogram of speech mixed with noise, there are always small time-frequency patches where the target speech has a higher signal-to-noise ratio than the background noise. In Glimpse-Based Metrics for Predicting Speech Intelligibility in Additive Noise Conditions, Yan Tang et al. showed that by simply counting these "glimpses" of clean audio energy, algorithms can accurately predict human speech intelligibility against stationary, non-speech noises.

Spectrogram Glimpsing Concept:
[Target Speech] + [Stationary Noise] -> Clear acoustic "glimpses" -> Brain reconstructs message (High Intelligibility)
[Target Speech] + [Competing Talker] -> Clear acoustic "glimpses" -> Brain gets confused by WHICH stream a glimpse belongs to (Low Intelligibility)

However, these glimpse-based metrics break down in speech-on-speech scenarios. As Yan Tang et al. later noted in A glimpse-based approach for predicting binaural intelligibility with single and multiple maskers in anechoic conditions, standard energy-counting models cannot account for informational masking.

When the masker is another person's voice, the listener still catches plenty of acoustic glimpses. The problem is that the brain cannot reliably determine which glimpse belongs to the target speaker and which belongs to the distractor. The failure occurs not because the signal is missing, but because the auditory scene analysis mechanism misallocates the incoming data.


Blurring the Boundary Between Ear and Brain

While it is conceptually useful to separate energetic masking (peripheral/physical) from informational masking (central/cognitive), the boundary between them is surprisingly fluid. Higher-level cognitive processes can actually change how our ears handle physical masking.

Consider an experiment by Maury Lander-Portnoy titled Release from Energetic Masking Caused by Repeated Patterns of Glimpsing Windows. The study introduced a masking noise that was spectrally identical to speech, creating heavy energetic masking. However, the author structured the noise so that its acoustic "glimpsing windows" repeated in a predictable temporal pattern.

Even though the physical energy overlapping the speech remained high, listeners showed a significant improvement in speech recognition when the noise pattern became predictable. This reveals that the brain uses top-down pattern recognition and auditory stream segregation to actively mitigate what was traditionally thought to be a purely peripheral, physical masking effect.

Ultimately, understanding speech in noise is a tightly integrated dance between the physical constraints of our ears and the cognitive architecture of our brains. We do not just hear with our cochleas; we filter, predict, and decode with our minds.

Key Foundational Work

Effect of number of masking talkers on speech-on-speech masking in Chinese (2007) — contributed 3 concept elements to this direction.
Modeling the perception of multitalker speech (2005) — contributed 2 concept elements to this direction.
Time-compressed speech perception with speech and noise maskers (2007) — contributed 2 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_informational_masking_09eed4ea,
  title = {informational masking},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/informational-masking/rdr_09eed4ea0b324fbf943364ce495ab082},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-22},
}

This exact version (version 1, frozen text):

@misc{lacuna_informational_masking_09eed4ea_v1,
  title = {informational masking},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/informational-masking/rdr_09eed4ea0b324fbf943364ce495ab082/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-22},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake