Chapter 7: Speech Perception: Auditory Processing and Language Sounds

Welcome to Last Minute Lecture.

This free chapter overview is designed to help students review and understand key concepts.

These summaries supplement not replaced the original textbook and may not be redistributed or resold.

For complete coverage, always consult the official text.

Welcome back to the Deep Dive.

I want to start today by asking you to do something very simple.

I want you to just pause for a second and notice exactly what is happening right now.

You are listening to a voice.

You aren't seeing me.

You aren't reading subtitles.

You're just, you're receiving this invisible stream of noise and somehow instantaneously, without even trying,

vivid images and complex ideas are popping into your head.

It feels like telepathy, doesn't it?

It really does.

It feels effortless.

It feels immediate.

But here's the thing, and this is what we're going to unpack today.

That feeling of effortlessness is, well, it's the biggest lie your brain ever told you.

That's a good way to put it.

If we actually pop the hood and look at the mechanics of what's happening, it's nothing short of a miracle.

We didn't think that speech just jumps from my mind into yours, like I'm handing you a package.

But the reality is so much messier.

Oh, absolutely.

As the source material for today puts it, speech is essentially a dance of air molecules.

That is such a beautiful poetic phrase for something that is physically quite chaotic, a dance of air molecules.

We take it for granted because we're so good at it.

We just assume that because we hear words clearly like cat or dog or philosophy, that the information is traveling to us in these neat, discreet little packages.

Right, like little text messages flying through the room.

Exactly.

But when you actually look at the physical signal, the raw physics of the sound wave hitting your ear, it is absolute chaos.

It isn't neat.

It isn't discreet.

It's a continuous, messy, overlapping stream of pressure changes.

The fact that you can extract meaning from that is arguably one of the most complex cognitive feats human beings perform.

And that is exactly where we're going today.

We're doing a deep dives into Chapter 7 of Cognitive Psychology, Classic Edition, which covers speech perception.

But this isn't just about hearing.

If you take one thing away from today, it should be this.

Listening isn't passive.

It is an act of creation.

Precisely.

This topic is absolutely pivotal for understanding the cognitive approach as a whole.

If you want to understand how the mind works, you have to understand speech perception because it illustrates the core argument of the entire field.

Our knowledge of the world is mediated.

It's not direct.

Mediated.

Let's dig into that word.

Well, think about it.

You don't just get the meaning of a sentence directly from the environment.

The meaning doesn't exist in the air pressure.

The meaning doesn't exist in the vibrating eardrum.

The meaning only exists after your mind actively constructs it.

You're rebuilding reality based on imperfect clues.

It's like the difference between looking through a clear glass window versus looking at a painting that your brain is hurriedly painting in real time based on some sketches it received from the ear.

That is a very apt analogy.

The ear provides the sketch.

The brain paints the masterpiece.

And sometimes, as we'll see, the brain paints things that aren't even there.

I love that.

So we have a huge journey ahead of us.

We're going to start with the physics, the actual raw data of sound, which is surprisingly gnarly.

Then we'll move into the biology, what the ear actually does with that data.

Then we get into the really tricky stuff, the building blocks of language, the puzzle of how we slice up sound into words when there are no spaces between them.

Which is a huge problem.

And finally, the major theories of how the brain pulls this magic trick off.

And we will see that the leading theory, something called analysis by synthesis, suggests we're doing a lot more work than we realize.

Okay, let's start at ground zero, the input.

Before it's a word, before it's a thought, what actually is sound?

When I speak, what am I physically doing to the room we're in?

At the most elementary level, sound is just a change in pressure.

It is a succession of compressions and expansions in the air.

When you speak, your vocal cords vibrate, and that pushes air molecules together, that's compression, and then pulls them apart.

Raffaction.

This creates a pressure wave that travels outward at roughly 700 miles per hour until it hits something.

In this case, the listener's eardrum.

Right.

It pushes the eardrum in and pulls it out.

That is it.

That is the raw data, time and pressure.

The source materials spend some time visualizing this, which I think is helpful because sound is invisible.

They describe the difference between a pure tone and what we actually hear in the real world.

Right.

Imagine a graph.

On the vertical axis, you have air pressure, high pressure at the top, low at the bottom.

On the horizontal axis, you have time.

If you strike a high -quality tuning fork, you get a pure tone.

On that graph, this looks like a sine wave, a perfect, smooth, regular, curvy line rolling up and down.

It's mathematically perfect.

That represents a single frequency.

But speech isn't a tuning fork.

If I graphed my voice right now, it wouldn't look like a smooth rolling hill.

No, absolutely not.

Speech or a piano note or a car horn produces what we call a complex wave.

If you graph that pressure over time, it looks like a jagged, chaotic mountain range.

It's spiky.

It's irregular.

It looks random.

But it isn't random, right?

Because my brain can tell the difference between me saying hello and a car honking.

Correct.

And here is the key insight from acoustics, which goes back to a mathematician named Fourier.

Any complex wave, no matter how jagged or messy looking, can be broken down into a sum of simple sine waves.

Wait, so the jagged mountain range is actually just a pile of smooth rolling hills stacked on top of each other.

Exactly.

It's like a soup.

You taste a complex soup and it has a distinct flavor.

But if you have a sensitive enough palate or a chemical analyzer, you can list the ingredients.

It has this much salt, this much tomato, this much basil.

The complex wave is the soup.

The simple sine waves are the ingredients.

And that list of ingredients, the frequencies and their loudness, is that's what we call the spectrum.

Yes.

The spectrum breaks the sound down.

It tells you, okay, this sound has a lot of low frequency energy, a little bit of middle, and a spike of high frequency.

So logically, this seems like the solution to speech perception.

We just take the sound, break it down into its spectrum, its ingredients, and then we match that recipe to a dictionary.

Oh, this recipe is the word hello.

Problem solved.

If only it were that simple.

This is where we run into the first major roadblock.

The problem with a standard spectrum is that it's a snapshot.

It shows you the ingredients at a single frozen instant of time.

But speech happens over time.

It's a dynamic event.

The text uses the word you as an example.

Y -O -U.

Let's analyze that.

Say the word you very slowly.

Yeah, yeah.

Feel how your mouth changes shape.

The sound quality is shifting constantly.

At the very beginning with the yeah sound, high frequencies predominate.

But as you slide into the ooh, the high frequencies drop out and low frequencies take over.

A static snapshot cannot capture that movement.

It's like trying to describe the plot of a movie by showing someone a single photograph of the main character.

You miss the action.

So we need a movie camera for sound.

We need to see how the ingredients change over time.

And that brings us to the spectrogram.

This seems to be the holy grail of speech analysis in the classic cognitive era.

It is the standard tool.

We need to visualize this carefully because it's a bit of a brain twister to convert sound to sight.

Imagine a 3D graph.

You have time, you have pitch, frequency, and you have loudness.

But 3D graphs are hard to read on paper, so we flatten it.

Okay, let's paint this picture for the listener.

Imagine a piece of paper.

Right.

The horizontal axis going left to right is time.

That's the timeline of the recording.

The vertical axis going up and down is frequency pitch.

High notes at the top, low bass notes at the bottom.

And how do we show loudness?

We use ink, darkness.

The darker the smudge on the paper, the louder the sound is at that pitch.

So silence would be a white paper.

A loud high -pitched scream would be a dark black streak across the top.

Got it.

So if we look at a spectrogram of the word U, what do we see?

You don't see a flat line.

You see a dark band of energy that starts high up on the left side.

That's the Y.

And then it angles sharply downward to the right, ending low.

That's the U.

It visually represents that slide from high to low.

It looks like a slide on a playground.

This sounds incredibly detailed.

And back in the day, scientists looked at these spectrograms, these visible speech patterns, and got really excited.

They thought, hey, we can just teach people to read these.

Or better yet, build a machine to read them.

That was the dream of the 1940s and 50s.

They called it the speech typewriter.

The idea was simple.

You speak into a microphone.

The machine makes a spectrogram.

It reads the visual patterns.

Oh, that slide is a U.

That blob is a hello.

And it types out the words.

And this is where the story takes a turn.

Because it didn't work.

It failed miserably.

And it failed for a reason that is absolutely central to our deep dive.

The visual patterns on the spectrogram are not consistent.

What do you mean?

Well, imagine I say, can you come?

And we make a spectrogram.

We get a nice picture.

Now imagine you say, can you come?

Your voice is deeper or you speak faster.

The picture looks completely different.

Now imagine I say it when I have a cold.

Or I shout it.

Or I whisper it.

The acoustic picture keeps changing.

Drastically.

The source material shows an example of five different spectrograms of the exact same phrase, can you come?

Spoken by the same person in different ways.

To the eye, they look like five completely different events.

One is long and stretched.

One is squished.

One has fuzzy bands.

Yet, if I played those recordings to you, you would hear the same words every single time.

So the machine looks at the data and sees chaos.

It sees five different things.

But the human ear looks at that chaos and hears, can you come every time?

Correct.

The machine struggles because there is no simple one -to -one mapping between the picture and the word.

This was the first big clue that speech perception isn't just about passively reading the acoustic signal.

If it were, the machine would work.

Since we can understand speech despite this massive variability, our brains must be doing something much more sophisticated than just template matching.

We're interpreting, not just recording.

But before we get to the interpretation, the software of the brain, we have to talk about the hardware.

The ear itself.

The biological interface.

We mentioned the eardrum, but the real magic happens in the cochlea.

This is that snail shell looking thing in the inner ear.

Right.

The cochlea is fluid filled.

And what's fascinating is that the cochlea basically acts as a biological spectrogram machine.

Inside that snail shell, there is a membrane, the basilar membrane.

It's stiff at one end and floppy at the other.

Like a piano string.

Sort of.

When sound hits it, different parts of the membrane vibrate depending on the frequency.

High frequencies vibrate the stiff end.

Low frequencies vibrate the floppy end.

So before the signal even reaches your brain, your ear has already performed a Fourier analysis.

It has already broken that complex soup down into its ingredients.

So we have a neural spectrogram going into the brain.

But then we run into the real world.

We rarely listen to clear speech in a soundproof booth.

We live in a noisy world.

The text talks about masking.

Right.

Masking is simply when background noise covers up the speech signal.

White noise, like a shower running or static, is very effective at this because it contains energy at all frequencies.

It blankets the speech.

But we're surprisingly good at dealing with this.

I mean, think about a crowded bar.

This leads us to the famous cocktail party phenomena.

I love this concept because we've all lived it.

You're at a party.

It's loud.

Music is thumping.

20 people are shouting.

But you can somehow lock in on one single conversation.

And you can ignore the others.

It's a superpower we don't appreciate.

But the question is, how?

How do we filter out the garbage?

The source material makes a really interesting point here.

It's not just about the meaning of the words or the pitch of the voice.

It's about location.

Location.

Auditory localization.

We have two ears.

And that is crucial.

This is called binaural hearing.

If a sound comes from your right side, it hits your right ear slightly sooner than your left ear.

And I mean slightly.

We are talking microseconds.

It's also slightly louder in the right ear because your head blocks some of the sound from reaching the left ear.

So my brain is basically doing high speed trigonometry in real time.

Signal A arrived 0 .0006 seconds before signal B.

Therefore, the source is 45 degrees to the right.

Essentially, yes.

And because you can locate the sound sources in 3D space, you can spatially separate the noise, the chatter on your left, from the signal, your friend on your right.

You can tell your brain, tune into the channel at 45 degrees right and mute everything else.

There's an experiment mentioned that proves this relies on having two ears.

If you record a cocktail party with a single microphone and then listen to that tape, it is a nightmare.

It's just a wall of noise.

You can't follow a conversation to save your life.

Because the single microphone flattened the 3D space.

It lost the timing differences.

Exactly.

Without the spatial cues, the brain can't separate the soup.

It just hears noise.

This shows us that hearing involves a massive amount of computational processing regarding space and time before we even get to the words.

That is wild.

It really shows how much processing is happening just to get a clean signal.

Yeah.

But okay, let's say we have the signal, we've tuned out the party.

Now we have to actually understand the words.

Yeah.

And this brings us to what I think is the most mind -bending part of the chapter.

The illusion of continuity.

This brings us back to looking at the spectrogram.

When we look at speech on paper or even the acoustic wave, it's a continuous stream.

It's a river of ink.

It flows.

But when we hear speech, we don't hear a river.

We hear bricks.

We hear beads on a string.

We hear discrete units.

We hear letters.

We hear words.

Exactly.

We hear segments.

But the acoustic signal doesn't have segments.

So where do they come from?

The text introduces the concept of the phone.

This is the atom of language, right?

The phoneme is the linguist's basic unit.

It's defined as the smallest unit of speech that changes meaning.

Think of the words lick, kick, and luck.

Lick, kick, luck.

The difference between lick and kick is the first sound.

L versus K.

The difference between lick and luck is the vowel.

Those tiny sound units that flip the definition of the word are phonemes.

But, and this is a huge, but a phoneme is a psychological category, not a specific sound.

What do you mean it's a category?

An L is an L, isn't it?

Not really.

Take the phone K.

The text gives the example of the words cool and keep.

I want you to whisper those two words to yourself right now.

Cool,

keep.

Pay attention to where your tongue hits the roof of your mouth.

Okay, cool, keep, cool, keep, whoa.

What do you feel?

When I say cool, the K is way back in my throat.

When I say keep, the K is pushed forward almost behind my teeth.

Exactly.

Acoustically, those are two different sounds.

If you put them on a spectrogram, they look different.

But your brain doesn't care.

Your brain says close enough, throw them both in the K bucket.

So a phone is like a bucket that we throw a bunch of similar sounds into.

Yes.

It denotes otherness.

It distinguishes one word from another.

In English, we don't care about the difference between the K and cool and keep.

But here's the kicker.

In other languages, like Arabic, that difference might change the meaning of a word.

So for an Arabic speaker, those would be two different phones.

They would hear them as distinct letters.

That is fascinating.

So our brain is trained to ignore certain acoustic differences and focus on others depending on the language we grew up with.

Precisely.

We are filtering the world through our native language.

Now, the text goes even deeper than phonemes.

It introduces the work of Jacobson and Halley on distinctive features.

This part felt very digital to me, like the Matrix.

It is very digital.

Their theory is that the foam isn't the bottom of the barrel.

Phonemes are actually bundles of features.

And these features are binary.

They're yes -no switches.

Like computer code.

Zeroes and ones.

Very much so.

Let's take the sound ZNS.

Zzzzzz.

Acoustically, they're very similar.

The mouth shape is almost identical.

The only difference is that for Z, your vocal cords are vibrating.

You can feel the buzz in your throat.

For S, they are not.

So in this binary system, Z is A plus voiced, plus voiced.

And S is A minus voiced.

Correct.

And there's a whole checklist of these.

Nasal versus oral is air coming out the nose.

Consonant versus vocalic.

The theory is that when sound hits your brain, you aren't hearing music.

You're running a checklist.

Is it voiced?

Yes.

Is it nasal?

No.

Is it continuous?

Yes.

And that unique combination of yes -no answers outputs the letter Z.

It sounds like a great system.

Very logical.

If the brain is a computer, this makes sense.

But I sense a however coming because nothing in this chapter is ever that simple.

You know me too well.

The problem is what's called the invariance problem.

If this theory were perfect, you would hope that every time there's a plus voiced feature, there is a specific unchanging acoustic signature in the sound wave.

A little red flag in the audio that says, I am voiced.

But there isn't.

No.

The acoustic cues for these features change depending on the context.

The features are defined by the words and the words are defined by the features.

It's a loop.

As the text says, we recognize features by words and words by features.

This theme of context just keeps coming up.

We can't just look at the raw data.

We have to look at what's around it.

This leads us to the problem of segmentation.

If the spectrogram is a continuous smear, how on earth do we know where one word ends and the next begins?

This is one of the most stubborn obstacles in speech perception.

We feel intuitively like there are little silences between words.

If I say the dog ran, you hear the space dog space ran.

But if you look at the acoustic signal, those spaces do not exist.

They're hallucinations.

They're cognitive constructions.

In fact, the text points out something that blows my mind every time.

You often pause longer inside a word than between words.

Wait, really?

Give me an example.

Take the phrase, his sliness made me suspicious.

Say it naturally.

His sliness made me suspicious.

Okay.

Think about the word sliness.

There's a distinct stop between the Y and the N.

Sliness.

You close your mouth.

Now think about made me.

Your looks are closed for the M in made and they stay closed for the M in me.

It flows together perfectly.

Acoustically, there is a bigger gap inside sliness than between made and me.

Yet I hear sliness as one block and made me as two blocks.

Exactly.

The silence isn't in the sound.

It's in the grammar.

That's why listening to a foreign language feels like a machine gun rapid fire stream of noise.

You've probably experienced this.

You turn on a French movie and it sounds like blah, blah, blah, blah.

You think, my God, they speak so fast.

How do they breathe?

Right.

But they aren't speaking faster than you.

You just don't know where to put the gaps.

You don't have the segmentation rules for French.

So if there are no physical gaps, what unit does the brain actually listen for?

Is it the phone,

the syllable, the phrase?

The text describes some really clever experiments to figure this out.

This is the great debate.

Let's look at the syllable argument.

There is some strong evidence that we process speech in syllable size chunks.

The tape speed experiment supports this, right?

Yes.

Huggins and Cherry did these experiments.

If you take a tape recording of speech and speed it up, does it become unintelligible?

Eventually, sure.

But we can handle a lot of speed.

I listen to podcasts at 1 .5x speed all the time.

Exactly.

And that is a problem for any theory that says we need a fixed amount of time,

say, exactly 100 milliseconds to process a phone.

If we speed up the tape, we're shrinking that time window.

Yet we still understand it.

This suggests we aren't processing in fixed time slots.

But then there's the interruption experiment.

This one confused me a bit at first, but it's really cool.

It is.

Imagine wearing headphones.

The speech switches back and forth between your left ear and your right ear.

Left, right, left, right.

If you switch it really fast, like 10 times a second, it sounds weird, but you can understand it.

If you switch it really slow, it's fine.

But at a specific rate, about three or four times per second, intelligibility drops to near zero.

You can't understand a thing.

Why that specific rate three times a second?

That rate corresponds to the average length of a syllable, about 0 .6 syllables per chunk.

The theory is that if you chop the speech stream exactly at the rate of syllables, you're destroying the integrity of the unit.

You're cutting the baby in half, so to speak.

The brain grabs a chunk, but it's half a syllable and it can't process it.

This strongly suggests the syllable is a key functional unit.

OK, so we have evidence for the syllable.

But then comes the heavy hitter, the phrase.

This brings us to the click experiments by Fodor, Bever, and Garrett.

This seems to be the smoking gun for the cognitive argument.

This is, in my opinion, the most critical section.

If you want to understand how active the brain is, you have to understand the click experiment.

The setup is simple.

Subjects listen to a sentence.

During the sentence, a click noise, like a static pop, is superimposed on the tape.

The subject simply has to say, where did the click happen?

Which word was it on?

Sounds easy.

I assume they're good at it.

They're terrible at it.

Listeners consistently migrate the click.

They don't hear it where it actually occurred.

They hear it at the nearest grammatical break.

If the click happens in the middle of a phrase, their brain literally picks it up and moves it to the gap between phrases to preserve the integrity of the unit.

That is wild.

But how do they know it's the grammar doing it, and not just pauses in the speaker's voice?

They designed the influence experiment to prove exactly that.

This is brilliant experimental design.

They recorded two sentences.

I want you to read them for us to hear the rhythm.

OK, here's sentence one.

As a direct result of their new invention's influence,

the company was given an award.

OK,

in that sentence, where is the natural grammatical break?

After influence.

As a result of the invention's influence,

pause.

The company was given an award.

Right, now read sentence two.

The retiring chairman, whose methods still greatly influenced the company, was given an award.

And the break there.

After company.

Whose methods influenced the company, pause, was given an award.

Correct.

The grammar puts the boundary in different places.

Now, here's the trick.

The researchers took the recording of the phrase influence the company, those three specific words, and they physically spliced the tape.

They used the exact same recording of those three words for both sentences.

So acoustically, they're identical.

Same speed, same intonation, same everything.

Identical.

And then they put a click right in the middle of that phrase.

And what happened?

In sentence one, where the grammatical break is after influence, listeners reported hearing the click earlier near influence.

In sentence two, where the break is after company, they reported hearing the click later near company.

That is mind -blowing.

The acoustic input was exactly the same.

The click was in the exact same spot, but the grammar moved the click in the listener's mind.

Precisely.

This proves that grammar determines how we hear.

Segmentation isn't just passive listening.

It is an active construction based on the rules of language.

We hear the sentence structure, not just the sound.

We are organizing the world to fit our rules.

So if perception is this active construction, how are we actually doing it?

We need a theory.

The text outlines three major approaches.

Yes, Licklider classified them.

First, you have correlation or templates.

This is the idea that you have a library of recordings in your head.

You hear a sound and you compare it to your library until you find a match.

Like Shazam.

Shazam listens to a song and matches it to a database.

Right.

But Shazam fails if you hum the song or sing it off key.

And that's why template theory fails for speech.

Speech is too variable.

If you have a different accent or pitch or speed, the template doesn't match.

You need an infinite library.

Okay, so that's out.

What's next?

Second is filtering.

This is the idea of feature detection.

You have a bank of filters.

Is it voiced?

Is it nasal?

And you identify the sound based on the features.

This is better, but it still struggles with that context problem we talked about.

Which leaves us with the third and most robust theory,

analysis by synthesis.

Analysis by synthesis.

It sounds technical, but it's actually very intuitive.

It means we analyze the input by synthesizing or creating a match internally.

Think of it as a guessing game.

Walk me through the steps.

I hear a word.

You hear a messy sound.

Your brain makes a quick, rough analysis.

Okay, sounds kind of like Apple.

Then your brain performs a magic trick.

It internally generates the neural signals for the word Apple.

It simulates saying Apple and checks that simulation against the sound you just heard.

If it matches, you understand.

So I'm not just receiving your speech.

I am actively rebuilding it in my own head to see if it fits.

Exactly.

It's a loop of guess, check, confirm.

It's active.

You are matching the input against a model you built yourself.

And it's connected to the motor theory of speech perception, right?

This was Lieberman's big idea.

And honestly, it sounds the most sci -fi of them all.

It does.

The motor theory is a specific, very physical version of analysis by synthesis.

Lieberman argued that we perceive speech by referencing how we would physically speak it.

We don't perceive the sound.

We perceive the gesture.

Wait, that sounds crazy.

You're saying I literally move my mouth or my brain moves my mouth.

When I listen to you, I'm sitting still right now.

You might think you're sitting still, but Lieberman would say your motor cortex is lighting up.

The best evidence for this is the D versus G experiment.

This is figure 37 in the text.

Lieberman looked at the spectrograms for the syllables D and do.

Okay, D and do.

Acoustically, the transition for the D sound looks completely different depending on the vowel.

In D, the frequency swoops up.

In do, it swoops down.

If you showed those two swoops to a physicist, they would say these are two completely different events.

Though the sound pattern is totally different.

Totally different.

Yet we hear the same letter, D, Y.

Lieberman asked, what is constant?

It's not the sound.

It's the tongue.

The tongue.

To say D, you tap the tip of your tongue behind your upper teeth, duh.

It doesn't matter if you're about to say E or ooh, that tongue tap is constant, Lieberman argued.

The sound changes, but the muscle movement is constant.

Therefore, since we hear a constant D, we must be listening to the muscle movement, not the sound.

So we decode speech by reverse engineering the motor commands required to make it.

That is efficient, actually.

It solves the invariance problem perfectly.

The acoustic signal varies, but the motor command is stable.

So we track the motor command.

But is there evidence that we're actually simulating muscle movements when we listen?

Or is it just a nice theory?

There is evidence.

Scientists have found subvocal activity.

When you listen intently to someone, or even when you're reading silently to yourself, there are tiny measurable electrical signals in your throat and larynx muscles.

Your vocal cords are twitching in sympathy with the words you are hearing.

That is incredibly spooky.

Reminds me of the hallucinations mentioned in the text.

Yes.

This is a dark side of the theory.

Schizophrenic patients who hear voices often show muscle activity in their vocal cords that matches what the voices are saying.

It suggests that their internal synthesis engine is running.

They're generating speech internally.

But they are failing to recognize it as their own.

They think it's coming from the outside.

But surely there's a hole in this motor theory.

I can think of one immediately.

What about people who can't speak?

That is the classic counter -argument.

Mutes.

There are people who cannot speak due to physical disability, maybe a stroke or a congenital issue, yet they can understand language perfectly.

If understanding required simulating muscle movement, they shouldn't be able to understand.

Right.

If I can't move the muscle, I shouldn't be able to run the simulation.

Exactly.

So that forces us to refine the theory.

We move to a more abstract version of analysis by synthesis proposed by Halley and Stevens.

Doesn't have to be muscular movement.

It can be a silent calculation.

We use our internal rules, our grammar, our vocabulary, our context to generate the guess.

We simulate the abstract structure, not necessarily the physical twitch.

And this brings us back to the power of context.

If we're guessing, then having clues helps us guess better.

The text mentions Miller's noise experiment.

This is a classic.

Miller found that if you play words in a noisy room, people identify them much better if the words are in a sentence.

For example, hearing, Don brought his black bread, is easier than hearing the jumbled list, Bread Black has brought Don, even if the noise level is exactly the same.

Because the sentence structure limits the possibilities.

It helps the synthesis engine make a better guess.

If I hear the dog wagged his, my brain's already guessing tail.

I'm synthesizing tail before you even say it.

Exactly.

It narrows the search space.

And this constructive nature creates illusions.

Have you ever heard of the verbal transformation effect?

I have.

That's the loop thing, right?

Yes.

Warren and Gregory discovered this.

If you take a tape loop of a single word, say rest, and play it over and over again, rest, rest, rest, rest, something weird happens.

You stop hearing rest.

Your brain gets bored.

The current hypothesis,

rest, stops satisfying the system.

So the synthesis engine starts trying out new organizations of the sound.

You might start hearing stress, or tress, or ester.

The acoustic input hasn't changed.

It is the same loop.

But your brain's hypothesis testing starts to drift.

Rest becomes stress.

That is wild.

It really proves that what we hear is a decision, not a fact.

It creates a hallucinatory quality.

The text even mentions a researcher, Dr.

Ditborne, who played a loop of a nonsense word all night while he slept to see what would happen.

Let me guess.

He went crazy.

He woke up hearing a completely different message that lasted for seven minutes.

His brain took the noise and spun an entire narrative out of it.

Okay, note to self.

Do not sleep with a tape loop playing.

But it drives the point home.

Listening is not passive.

It is an active, creative, and sometimes slightly unstable process.

It is.

The text concludes by distinguishing between the passive and active modes.

There is a pre -attentive phase.

The cochlea analyzing frequencies may be spotting some features.

But then there's the active phase analysis by synthesis.

We take those fragments and we build a world out of them.

So let's recap where we've been.

We started with the physics, the sine waves, and the messy complex waves.

We looked at the spectrogram and realized there is no simple roadmap in the sound itself.

It's a mess of variables.

We talked about the biological tools, the cochlea and the two ears that let us solve the cocktail party problem by calculating geometry in microseconds.

We looked for the units of speech phones, syllables, and found that none of them are perfect, fixed, distinct things in the audio.

They are categories our brains impose on the sound.

And finally, we realized the only way to explain how we segment words and understand sentences, especially with those click experiments, is to accept that we are constructing the meaning internally.

We're using grammar and context to synthesize a guess that matches the input.

We don't just hear what is there.

We hear what we expect to be there based on the rules we know.

We are architects of our own auditory reality.

It changes how I think about conversation.

Every time I talk to someone, I'm not just receiving data.

I'm actively participating in a simulation.

It raises a fascinating philosophical question, doesn't it?

If the leading theory is true, if we understand speech by internally simulating the production of that speech.

Here is where it gets really interesting.

Does that mean that every time you listen to a friend, you are, on a subconscious level, acting out their part?

Are you acting as them in your own mind just to understand them?

That is a provocative thought.

We are all just secretly mimicking each other in the privacy of our own skulls.

In a way, yes.

Empathy might be built into the very mechanics of perception.

To understand you, I have to become you, just for a millisecond.

I love that.

It makes listening feel like a much more connected act.

Well, that wraps up our deep dive into speech perception.

A huge thank you to the Last Minute Lecture team for the source material.

We hope you'll never listen to a conversation the same way again.

Until next time, keep listening actively.