011 min
Cocktail Party Effect at a glance
- What it is: Listeners can pick out one talker in a crowd. A few of them also notice their own name in a voice they were not following.
- Origin: Colin Cherry named the cocktail party problem in 1953. Neville Moray showed the name effect in 1959.
- Where it bites: A message that contains something personal can pull a person away from the task they are doing.
- Guard against it: Give simultaneous audio different voices and different positions, and keep personal alerts out of focused work.
- Limit: Only about one third of listeners notice their own name in the ignored voice.
021 min
Where the Cocktail Party Effect shows up
Picture a designer who is in an open-plan office, writing up the findings of a usability test. Four conversations are going on around them, and they hear none of the words. They are reading and typing, and the talk is a general hum.
Then a colleague three desks away says the designer's name while talking about something else. The designer stops typing and looks up. A moment earlier they could not have repeated a single word of that conversation.
Now change one input. The colleague says a different name, the name of a person the designer has never met. The designer keeps typing. The sound is the same in every way except one: the name now means nothing to them.
This pattern has two parts. A listener can follow one voice and ignore the rest. And ignored speech is not fully ignored, because a few meaningful items, such as a listener's own name, can still reach awareness. Researchers use the term cocktail party effect for both parts. Sometimes the label means only the second part, noticing your name. Both are described below.
032 min
Why the Cocktail Party Effect happens
The cocktail party effect has two linked mechanisms. The first is how a listener separates one voice from a mixture. The second is how an ignored voice can still interrupt.
Separating one voice from a mixture
Two voices speaking at once arrive at each ear as one combined sound wave. The auditory system has to split that wave into separate sources. This is called auditory scene analysis: grouping the parts of a sound that belong together and keeping them apart from other sounds.
Sounds can be grouped using simple features. Two of the strongest are the direction a sound comes from and its fundamental frequency. The fundamental frequency is the rate at which the vocal folds vibrate, and a listener hears it as the pitch of a voice. Attention then selects one group and gives it more processing. This is selective attention, the process that chooses some input to process in depth and leaves other input with less.
The brain stores this selection. In a 2012 study, Nima Mesgarani and Edward Chang recorded signals from the surface of the cortex in people listening to two speakers at once. The recordings showed that the response in the auditory cortex represented the attended speaker, not the mix. A listener's brain activity, in other words, matched one voice and not the sum of both.
Why an ignored voice can still interrupt
Ignored speech is processed to some degree. A word that is highly relevant to the listener, such as their own name, can shift attention to the ignored voice for a moment. Researchers describe this as an involuntary shift of attention. It does not depend on the listener deciding to listen.
The adaptive reason is easy to state. A person who hears their name may be needed, so the system pays a small cost to check. The price is a brief loss of the task that was in progress. People who notice their name make more mistakes in the words right after it, as the studies below show.
The exact mechanism is still debated. Researchers disagree about how deeply unattended speech is processed. The section on competing theories explains the three main positions.
042 min
Where the Cocktail Party Effect comes from
The phrase comes from Colin Cherry, a British researcher who worked on communication engineering. In 1953 he asked what one person is saying when others are speaking at the same time, and he called this the "cocktail party problem". He wanted to know how a machine could do what listeners do. His paper, "Some experiments on the recognition of speech, with one and with two ears", appeared in the Journal of the Acoustical Society of America.
Cherry developed the dichotic listening procedure. A listener wears headphones and hears a different message in each ear. The listener must repeat aloud, word by word, the message in one ear. This task is called shadowing. People did it well. But when asked afterwards what the ignored ear had said, they could report only physical features, such as the gender of the voice or whether it was speech or a tone. They could report little or no meaning.
In one of Cherry's tests the ignored message changed from English to German halfway through. The listener reported that they did not know the language but assumed it was English. A change from a male voice to a female voice was nearly always noticed, and a pure tone was always noticed. From these tests Cherry concluded that detailed aspects of the ignored message, such as its language, individual words, or meaning, go unnoticed.
Donald Broadbent used results like these in 1958 to build a model of selective attention. The model said ignored input is removed early, using only its physical features.
In 1959 Neville Moray tested one gap in that model. He put the listener's own name into the ignored ear. Some listeners noticed it.
052 min
Who notices their own name
Moray found that only 33% of subjects reported hearing their own name when it was inserted into the ignored message. Popular accounts often say everyone notices their name at a party. The data say otherwise.
Wood and Cowan (1995) repeated the experiment with more precise equipment. They found 34.6% of subjects reported hearing their own name presented in the ignored message. They also recorded the shadowing. The listeners who noticed their name made more shadowing errors in the seconds that followed. Noticing the name cost them part of their performance on the main task.
Andrew Conway, Nelson Cowan and Michael Bunting asked in 2001 why some listeners notice and others do not. They tested 40 undergraduates and sorted them by working memory capacity. Working memory is the system that holds and controls the information a person is using right now, and its capacity differs between people. The researchers measured it with the operation span task, in which a person solves a simple sum and also remembers a word, over series of increasing length.
The result surprised many readers. One could expect that people with more capacity can monitor the ignored voice as well. The opposite happened. 20% of high-span subjects and 65% of low-span subjects reported hearing their name in the unattended message. Low-span subjects also made more shadowing errors overall, with an average of 20.88 against 10.00 for high-span subjects.
The authors interpret this as a difference in blocking out distraction. People with higher working memory capacity are better at keeping ignored information out. They are therefore less likely to hear their name and less likely to be disrupted by it. Your working memory capacity may help decide whether the effect happens to you.
061 min
Competing theories of the Cocktail Party Effect
Cherry's results started a line of work on how attention chooses what to process. Three positions came out of it. They differ on one question: how much of the ignored speech is processed?
| Theory | Claim | Where selection happens |
|---|---|---|
| Early selection (Broadbent, 1958) | Only physical features of ignored sound are processed. | Before meaning is analysed |
| Attenuation (Treisman, 1964) | Ignored input is processed up to meaning, but more weakly and more slowly. | Gradually, by reducing the strength of ignored input |
| Late selection (Deutsch and Deutsch, 1963) | All input is processed up to meaning. | After meaning is analysed |
The name effect looked like a problem for early selection. If only physical features of ignored speech are processed, nobody should notice a name. Attenuation explains it: a name is so familiar that even a weakened signal is enough to identify it.
Later researchers questioned the early studies themselves. Bronkhorst reviewed the field in 2015 and wrote that it is now clear that such processing can take place preattentively, meaning before attention is directed to it. He also noted that the depth of processing depends on how relevant the sound is to the task. So the question has moved from whether ignored speech is processed to how much, and when. The review lists many unresolved issues.
072 min
How the Cocktail Party Effect shows up in product and engineering
Designing simultaneous audio
A voice interface designer who plays two audio streams at once, such as a navigation prompt and a call, can use what listeners use. Bronkhorst's review finds that sounds can be grouped, and then selected, using primitive features such as spatial location and fundamental frequency. Two streams that differ in voice and position are easier to separate. Two streams in the same voice and place are not. This is a design inference from that finding, not a tested rule for every product.
Personal alerts and focused work
The name effect has a direct use in notifications. An alert that contains a user's own name or another strongly personal word is more likely to shift attention, and it also interrupts the task. Conway and colleagues found that a name briefly disrupted shadowing in the following words. Used on purpose, this is a reason to put a person's name in an urgent message. Used by accident, it is a reason not to put a name in a routine message. Notice the visual version too: banner blindness is the habit of ignoring page areas that look like adverts, while a personally relevant item can still catch the eye.
Speech separation in machine learning
Machine learning engineers work on the cocktail party problem as a software problem. John Hershey and colleagues described a method called deep clustering in 2015. A neural network turns each part of a spectrogram, a picture of sound frequencies over time, into a vector, and parts of the same speaker end up close together. On mixtures of two speakers it had not heard in training, the model improved signal quality by around 6 dB. It also generalised to mixtures of three speakers although it was trained only on pairs. Voice assistants, meeting transcription and hearing aids all face this problem. The result is a research benchmark, not a statement about any particular product.
081 min
How to guard against the Cocktail Party Effect
The effect is not an error to remove. It is a trade-off between staying on task and noticing relevant information. These steps change the conditions so the trade-off works for you.
Reduce speech you do not need.
Speech is the sound most likely to be processed for meaning. A review by Bronkhorst notes that task-irrelevant sounds impair performance on memory tasks, the irrelevant sound effect, and that this may come from involuntary shifts of attention. Close a door, move a seat, or use a room without talk when the work depends on memory.
Separate the voices you must listen to.
In a meeting with several remote speakers, put them in different positions in stereo or spatial audio where the tool allows it. Different positions are among the cues listeners use to separate voices.
Control what carries your name.
For a team, route personal alerts to a channel people expect to read. For a product, keep names for messages that need action.
Do not assume willpower fixes it.
Conway and colleagues found that the ability to block ignored speech tracked working memory capacity. That is a stable difference between people, so a rule such as "just ignore it" will work unevenly.
091 min
Cocktail Party Effect vs. nearby concepts
The effect is often mixed up with other attention effects. Each one differs on a single axis.
| Concept | What it is | Axis that separates it |
|---|---|---|
| This pageCocktail party effect | ||
| PsychologyInattentional blindness | ||
| PsychologyDivided attention | ||
| Not in the library yetIrrelevant sound effect |
Inattentional blindness is close in cause, since both come from attention being directed to one thing. The difference is the result. In the cocktail party effect a relevant item gets through. In inattentional blindness a visible item does not.
101 min
Common misunderstandings about the Cocktail Party Effect
"Everyone hears their name across the room." The original data do not say this. Moray found 33% of subjects reported hearing their name, and Wood and Cowan found 34.6%. The group that noticed was a minority in both studies.
"It proves we process everything we ignore." It does not. Cherry found that listeners could not report the language or the words of the ignored ear. Only a few highly relevant items, such as a name, appear to get through.
"It is a skill you train." The 2001 result links noticing to low working memory capacity. Hearing your name is therefore not a sign of better listening. It is a sign that distracting information was blocked less well in that test.
The effect is also useful. Noticing your name is how a person on a call notices that they have been asked a question. A fully closed attention system would miss it, so trying to remove the effect entirely would be a mistake.
111 min
Where the evidence is contested
The early studies have a design problem. Bronkhorst writes that possible shifts in attention to the ignored ear were at best poorly controlled, and the effects were not monitored. A listener who briefly looks at the other ear would hear the name without any special processing of ignored speech. Cowan and Wood addressed this by analysing the shadowing output and its timing, and the shadowing errors after a name show that attention did move.
Researchers also disagree about the cause. Broadbent, Treisman and Deutsch and Deutsch each explain the name effect differently, and no study has settled which is correct. Bronkhorst notes that the field needs models that are neurophysiologically plausible, and that attention to dynamically changing sounds is slow in a way no model yet explains.
The name effect itself has been replicated, in 1995 and 2001, with an effect of roughly one third of listeners. The sample sizes were small. The 2001 study used 40 subjects with 20 per group. A practitioner should treat the numbers as an indication of direction and not as a precise rate for any given workplace.
For design work, the contested part is the mechanism, not the behaviour. A personal word can interrupt a task for some people. That is well supported. Why it does so, and how often, depends on the person and the task.
122 min
Frequently asked questions about the Cocktail Party Effect
What is the cocktail party effect in simple terms?
It is the ability to follow one conversation in a room full of voices. It also refers to noticing your own name in a conversation you were not following.
What causes the cocktail party effect?
The auditory system groups sounds by features such as direction and pitch. Attention then selects one group. Highly relevant items in the ignored group can still shift attention.
Who discovered the cocktail party effect?
Colin Cherry described the cocktail party problem in 1953. Neville Moray showed in 1959 that some listeners notice their own name in an ignored message.
Does everyone hear their name in a crowd?
No. Moray found that 33% of listeners reported their name. A 1995 replication found 34.6%.
Does working memory affect the cocktail party effect?
Yes. In a 2001 study, 65% of low-span listeners and 20% of high-span listeners reported hearing their name in the ignored message.
How can machines solve the cocktail party problem?
Neural networks can separate mixed speakers. Deep clustering improved signal quality by about 6 dB on mixtures of two unseen speakers in a 2015 paper.
Is the cocktail party effect the same as selective attention?
No. Selective attention is the general process of choosing some input. The cocktail party effect is a specific case of it involving speech, and it includes the finding that some ignored items get through.
?7 questions
Questions people ask
What is the cocktail party effect in simple terms?
What causes the cocktail party effect?
Who discovered the cocktail party effect?
Does everyone hear their name in a crowd?
Does working memory affect the cocktail party effect?
How can machines solve the cocktail party problem?
Is the cocktail party effect the same as selective attention?
§6 sources
Sources
Show all 6 sourcesHide sources
Cherry, E. C. (1953). Some experiments on the recognition of speech, with one and with two ears. Journal of the Acoustical Society of America, 25(5), 975-979.
Moray, N. (1959). Attention in dichotic listening: Affective cues and the influence of instructions. Quarterly Journal of Experimental Psychology, 11(1), 56-60.
Conway, A. R. A., Cowan, N., and Bunting, M. F. (2001). The cocktail party phenomenon revisited: The importance of working memory capacity. Psychonomic Bulletin and Review, 8(2), 331-335.
Bronkhorst, A. W. (2015). The cocktail-party problem revisited: early processing and selection of multi-talker speech. Attention, Perception, and Psychophysics, 77(5), 1465-1487.
Mesgarani, N., and Chang, E. F. (2012). Selective cortical representation of attended speaker in multi-talker speech perception. Nature, 485, 233-236.
Hershey, J. R., Chen, Z., Le Roux, J., and Watanabe, S. (2015). Deep clustering: Discriminative embeddings for segmentation and separation. arXiv:1508.04306.





