Why the Same Sound Can Be Two Different Words
Ambiguous speech shows how the brain combines unchanged acoustic evidence with contextual expectations, and why degraded input can make expectation more influential rather than less.
A complete interactive classroom, not just a preview.
Start when you are ready to enter this Stage's 11 scenes and explore, respond, and learn as you go.
Why can one identical sound be heard as two different words?
- acoustic-identity
- A target sound can have a byte-for-byte identical waveform and spectrogram in two contexts.
- ambiguous-speech-perception
- Listeners can perceive the same ambiguous target as different words.
- polyphonic-perception
- Polyphonic perception describes hearing one sound as multiple possible linguistic items depending on context.
- top-down-predictive-processing
- Top-down processing and predictive processing use prior knowledge and context to constrain the interpretation of incoming sound.
- degradation-effect
- Noise and low-pass filtering can reduce acoustic evidence and strengthen contextual expectations.
- phonemic-restoration
- Phonemic restoration is the perceptual filling in of speech that is missing or masked, even though its effect depends on the acoustic context.
Two words sound different because their target waveforms or spectrograms are different.
Show that the target sound remains acoustically identical while only the preceding context changes.
More noise always makes two words harder to tell apart.
Demonstrate that degradation can make the intended word easier to identify by strengthening top-down expectations.
The brain simply hears all the acoustic detail present in the signal.
Show that perception is an active interpretation constrained by predictions rather than a passive copy of the waveform.
Phonemic restoration means the brain restores every missing sound equally well.
Explore how context determines whether a masked or missing speech segment can be perceptually restored.
- Basic understanding of waveform and spectrogram
- Awareness that speech sounds carry linguistic meaning
- detailed auditory neuroanatomy
- quantitative signal-processing derivations
- clinical disorders of speech perception
- cross-linguistic phoneme inventories
- Identify that the target waveform and spectrogram remain unchanged across contextual conditions.
- Predict how context changes the perceived word.
- Explain why degraded input can increase the influence of expectation.
- Connect the experiment to top-down and predictive processing.
- Distinguish phonemic restoration from literal acoustic recovery.
- Use the expectation-versus-evidence principle to explain ambiguous perception in another everyday listening situation.
Beginners who understand basic waveforms and spectrograms but have not studied speech perception.
- 01One Sound, Two WordsslideOrientationPredict
Pose the central paradox: the target sound in “naked” and “day and” is acoustically identical, yet listeners report different words.
- The puzzle
- What changes—and what does not
- Course map
- 02Hold the Signal FixedinteractivePredictionPredict
Change only the preceding context, inspect an invariant waveform and spectrogram, and predict which word will be heard before playing the clip.
- Context changes
- Target signal stays fixed
- Commit to a prediction
- 03Degrade the EvidenceinteractiveMisconception repairPredict
Add noise and low-pass filtering, then test the counterintuitive prediction that a blurrier target can make the contextually expected word easier to recognize.
- Noise and low-pass filtering
- Less acoustic evidence
- Stronger contextual pull
- 04The Brain as PredictorslideModel buildingExplain
Introduce top-down processing and predictive processing as a framework in which the brain constrains incoming evidence using prior knowledge and context.
- Bottom-up evidence
- Top-down expectations
- Perception as constrained interpretation
- 05Prediction Meets EvidenceinteractiveModel buildingConstruct
Explore the experiment from physical signal through sensory evidence, prediction, and perception, with animated reveals connecting each stage.
- Signal remains fixed
- Context changes predictions
- Interpretation changes
- 06Ambiguity Signal LabinteractivePracticeChoose
Tune context and degradation to maximize identification of the contextually expected word while the target signal remains unchanged.
- Adjust evidence quality
- Track identification accuracy
- Find the strongest expectation effect
- 07Polyphony and Illusory CompletionslideModel buildingExplain
Clarify polyphonic perception and introduce phonemic restoration as related examples of listeners using linguistic knowledge to organize incomplete or ambiguous input.
- Polyphonic perception
- Ambiguous speech perception
- Phonemic restoration
- Perceptual fill-in is not acoustic recovery
- 08Restore the Missing SoundinteractiveMisconception repairConstruct
Replace a masked or missing speech segment with noise and compare compatible versus incompatible contexts, observing when the brain restores a phoneme perceptually.
- Mask the target
- Switch context
- Hear perceptual completion
- 09Noisy Restaurant ChallengeinteractiveApplicationApply
Guide a listener through a noisy social scene by collecting context clues, then use those predictions to identify garbled messages and recover useful meaning.
- Gather contextual clues
- Update predictions
- Decode garbled speech
- 10Evidence or Expectation?interactiveAssessmentApply
Diagnose unfamiliar listening scenarios by adjusting signal quality and context, then explain whether evidence or expectation dominates each outcome.
- Compare competing explanations
- Explain counterintuitive results
- Transfer the principle
- 11The Gap Is EverythingslideSynthesisExplain
Synthesize the class around the distinction between acoustic evidence and perceptual expectation, including why blurrier input can reveal the brain's predictive machinery.
- The sound does not change
- Your expectation does
- That gap is everything
Discussion threads for a Stage aren't available yet.
