Transcript (1) limits of hearing

Hearing
Outline
(1) limits of hearing
(2) loudness
(3) pitch
(4) perceptual organization
(5) localization
(6) speech perception
(1) Limits of hearing
a) audibility function
b) detection of complex stimuli
c) masking & the critical band
d) patterns of hearing loss
range of hearing is 20 - 20,000 Hz
How good is the ear at detecting sounds?
How GOOD is the ear at detecting sounds?
We can present complex stimuli by breaking up pure tones.
Acoustic power can be
concentrated into a single
frequency (i.e., a pure tone).
P
O
W
E
R
Or distributed among multiple
frequencies presented
simultaneously (i.e., a complex
tone).
bandwidth
low
high
Frequency (Hz)
In addition, we can vary the
bandwidth (i.e., the difference
between the highest and lowest
frequencies) in the complex tone.
How does detectability of a sound depend on complexity?
(1) when frequencies are close together (narrow bandwidth)

detectability depends on the total acoustic power in the stimulus

the acoustic power is summed across frequencies.
(2) when the frequencies are far apart (wide bandwidth)

detectability declines when the frequency components in a
complex tone are too far apart

the auditory system is no longer capable of combining their
acoustic power.
The bandwidth at which the integration begins to fail is called the
critical bandwidth.
Detection of Complex Stimuli
nearby frequencies are combined
% detection
close together
far apart
critical
bandwidth
stimulus bandwidth
frequencies falling within a critical bandwidth are integrated
by the auditory system
Noise Masking
(1) broad-band noise contains all audible frequencies (“white noise”)
(2) band-pass noise contains a smaller range of frequencies
low frequency
cut-off
Power
high frequency
cut-off
centre frequency
Frequency (Hz)
(3) bandwidth is the difference between the high and low frequency
cutoffs.
Varying Bandwidth
Difference of high and low cut-offs
same centre frequency, different bandwidths
What does noise sound like?
depends on bandwidth
(1) a very narrow band noise sounds like a tone (centre frequency)
(2) white noise is a hissing sound (no pitch)
Detecting a signal embedded in noise
How does noise affect pure tone thresholds?
Consider a task in which you are asked to detect a pure tone (the
signal) embedded in a background noise.
(1) The noise is band-pass: the centre frequency equals the
signal's frequency.
(2) The bandwidth of the noise varies across experimental
conditions.
How does detectability vary with noise bandwidth?
Noise Band-pass: Varying bandwidths
Signal (pure tone)
P
O
W
E
R
Frequency (Hz)
Detecting a signal embedded in noise
Threshold
How does noise affect pure tone thresholds?
critical
bandwidth
no
noise
Noise Bandwidth
threshold increases with noise bandwidth up to a point, then levels off
Detecting a signal embedded in noise
How does noise affect pure tone thresholds?
(1) Increasing bandwidth does increase detection threshold.
•
this increase in threshold is referred to as masking.
•
we say that the signal is masked by the noise.
(2) Increasing noise bandwidth results in more masking, but only up
to a point.
•
beyond some critical bandwidth, increasing noise bandwidth
does not result in more masking.
•
The amount of masking levels off and thresholds are
constant.
Interpretation of noise masking
The critical band usually is interpreted as the width of an
frequency-selective auditory channel/filter.
• detect signal by monitoring response of 1 channel
• noise falling within channel masks signal
• noise falling outside channel has no effect
(1) This filter responds well to a small range of auditory frequencies.
(2) Analogous to the spatial frequency channels.
Frequency-selective filter
P
O
W
E
R
Frequency (Hz)
Frequency-selective filters (or channels)
(1) The assumption is that observers have many auditory
channels tuned to different frequencies.
(2) Some channels respond to low frequencies, some to high
frequencies, some to medium frequencies.
(3) In all cases, the auditory filter responds to a small range
around a preferred, or best frequency.
(4) Generally, the critical band increases with increasing
frequency.
Hearing Loss
Two types:
(a) conduction loss
(b) sensory/neural loss
Conduction hearing loss
Disorder of outer and/or middle ear
• problem associated with mechanical transmission of sound into cochlea
Causes:
• blockage of outer ear
• punctured eardrum
• middle ear infection
• otosclerosis
Effects:
• broad-band sensitivity loss
• conduction loss can be severe (up to 30 dB)
• loss occurs across a wide range of frequencies
• does not affect neural mechanisms
• treated with hearing aids/surgery (stapedectomy)
Sensory/Neural Loss
Damage to the cochlea and auditory nerves
• age-related hearing loss (Presbycusis)
• progressive sensitivity loss at high frequencies
• high frequency cutoff drops from 15,000 Hz (30
years) to 6,000 Hz (70 years)
Effects of Exposure to Noise
(1) Temporary Thresholds Shifts (TTS)
• short term exposure to noise (nightclub, concert)
• increases your thresholds about 30 dB
• recover over a period of several hours/days
(2) Permanent Thresholds Shifts (PTS)
• long term exposure to noise (work-related)
• short, intense sounds (explosion)
• can be as large as 60 dB
 (1) limits of hearing
(2) loudness
(3) pitch
(4) perceptual organization
(5) localization
(6) speech perception
Loudness
Loudness is a psychological, not physical,
attribute of sound!
• loudness is not the same as intensity (it is your
subjective perception of intensity)
• cannot measure loudness with a sound meter
How can we estimate loudness?
(1) Loudness Matching Method
(2) Magnitude Estimation Method
Loudness Matching Method
Task: Match the loudness of 2 tones.
(1) present a standard tone at a fixed intensity
(2) present a test tone at a different intensity
(3) adjust test's intensity until it has the same loudness as the
standard
(4) repeat procedure for many test frequencies
(5) graph of intensities at all test frequencies
(6) repeat entire procedure for a new standard intensity
Equal Loudness Contour
every point on the contour has the same loudness as the standard
Equal Loudness Contours
• When the standard intensity is low, equal loudness contours have
the same shape as the audibility function.
• When the standard intensity is high, equal loudness contours are
much flatter than the audibility function.
Another way of stating the results:
• At low intensities tones have the same loudness when they are
equally detectable
• At high intensities they match in loudness when they have the
same intensity.
Magnitude Estimation
In a magnitude estimation task, an observer rates the loudness
of a stimulus on a numerical scale.
(We are really good at this!)
The loudness ratings are related to intensity according by the
formula:
Loudness = k * Intensity0.67
(Steven’s Power Law)
Loudness
Loudness grows more slowly than intensity!
Intensity (dB)
Effects of noise mask on loudness
Consider a task in which observers rate the loudness of a pure tone
signal embedded in a band-pass noise centred on the signal
frequency.
Noise will mask the target, so we would expect to find that loudness
ratings are decreased for low intensity signals.
Q: What happens at higher intensities, when the signal is above
threshold?
A: Once above threshold, loudness increases rapidly -- much more
rapidly than normal.
Effects of noise mask on loudness
This figure shows loudness
judgements for a 1,000 Hz
tone presented in quiet (left
curve) and in 2 different
amounts of noise (middle
and right curves). Note how
all the curves converge at
high intensities. This
abnormally rapid growth
in loudness with increasing
intensity is referred to as
loudness recruitment.
Effects of hearing impairment on loudness
Some forms of hearing loss
(cochlear defects) affect
loudness judgements in
much the same way as a
masking noise: loudness for
weak, but not intense, tones
is reduced.
This phenomenon is another
example of loudness
recruitment.
Loudness of Complex Tones
complex tones vs. pure tones
Complex tones are sometimes louder than pure tones of equal
energy. Differences depend on the bandwidth of the complex
stimulus.
Zwicker, Flottorp, & Stevens (1957)
 compared loudness of pure tone (1,000 Hz) vs. complex tone (made
up of frequencies ranging from 900 to 1,100 Hz)
 listeners adjusted the intensity of the pure tone to match loudness
of complex tone
 experimenters varied the bandwidth of the complex tone
 all complex tones had equal energy
Effect of bandwidth on loudness
loudness changes at critical bandwidth
Intensity of
matching
tone
160 Hz
Stimulus Bandwidth
 (1) limits of hearing
 (2) loudness
(3) pitch
(4) perceptual organization
(5) localization
(6) speech perception
Pitch
Pitch is also a psychological, not physical, attribute of sound!
• pitch is not the same thing as frequency (it is your
subjective perception of frequency!)
• low frequencies (e.g., 500 Hz) will sound low in pitch
• high frequencies (e.g., 1500 Hz) will sound high in pitch
Place Theory of Pitch (of pure tones)
• peak activity is localized on
basilar membrane
• different frequencies => different
locations
• pitch is encoded by the place of
activation
Pros
(1) damaging parts of basilar membrane results in frequencyspecific deficits
(2) stimulating parts of basilar membrane evokes different pitches
Cons
(1) low frequencies have a pitch, but don’t produce localized
activity on basilar membrane
(2) Can’t account for pitch of complex sounds
Frequency Theory of Pitch
• pitch depends on rate, not place, of response
• neurons can fire in-synch with signals
• rate of response might represent frequency of stimulus
But….
cells can’t track high frequency sounds
Maybe there is a dual coding theory of pitch!
(place theory for high frequencies, response rate for low
frequencies)
Pitch of Complex Sounds
• most sounds are complex
• not all complex sounds have a pitch (white noise)
• those that do tend to have a harmonic structure
Pure Tone
low
Complex Sound (with pitch)
high
low
high
Frequency (Hz)
pitch of the complex stimulus = pitch of fundamental frequency
Fundamental Frequency
The fundamental frequency of a set of frequencies is the highest
common divisor of the set.
What is the fundamental frequency for each of the following sets?
200, 400, 600, 800, 1000 Hz
200, 600, 800, 1000 Hz
100, 400, 600, 800, 1000 Hz
200, 300, 600, 800, 1000 Hz
Each of the above sets has one fundamental frequency, although the
fundamental itself may not actually be present in the set.
Problem of the Missing Fundamental
(Virtual Pitch)
Is the fundamental really missing?
When the fundamental is missing from the stimulus, do
distortions produced by the ear introduce a fundamental
frequency?
Is the fundamental really missing?
effect of masking on virtual pitch
Intensity (dB)
band-pass mask centred on the
fundamental frequency
100
200 300 400 600
800 1000
Frequency (Hz)
masking does not eliminate the pitch!
 (1) limits of hearing
 (2) loudness
 (3) pitch
(4) perceptual organization
(5) localization
(6) speech perception
Perceptual Organization
How do you group sounds to identify an object?
Auditory Scene Analysis
Gestalt grouping principles apply!
(1) Good continuation: spectral harmonics
• harmonics are grouped together
• doe-rae-mi-fa-so-la-ti-doe
(2) Similarity: common spectral content
• similar frequencies are grouped together
(3) Proximity: common time course
• frequencies that occur at the same time are grouped
together
(1) limits of hearing
(2) loudness
 (3) pitch
 (4) perceptual organization
(5) localization
(6) speech perception
Cone of Confusion
Binaural cues do not provide unambiguous information about a
sound's location.
For example, the same binaural cues are produced by sounds
directly in front of and behind an observer.
A particular combination of IIDs and ITDs can be produced by
sounds in many different directions.
All such directions lie
on the surface of the socalled cone of confusion.
Minimizing Mislocalizations
Even though binaural cues are ambiguous, we don't make very
many localization errors.
How does the auditory system minimize mislocalizations?
(1) Head movements
•
Only one sound source location is consistent with
multiple head positions.
(2) Pinna
(3) Visual cues
Cocktail Party Effect
You can focus on one sound in a noisy environment.
This form of "sound segregation" is due in part to being able to tell
from where the sound is coming.
Binaural unmasking
• masking reduced when sounds come from different locations
Example of how binaural cues reduce masking:
(1) Present signal with noise over headphones to the left ear
•
you can't hear signal
(2) now add identical noise (via headphones) to the right ear
•
even though there is more noise, you now can hear the signal!
•
binaural unmasking occurs because the noise and signal have
different perceived locations
 (1) limits of hearing
 (2) loudness
 (3) pitch
 (4) perceptual organization
 (5) localization
(6) speech perception
Speech Perception
Speech perception different from perception of other auditory
stimuli.
Evidence:
(1) Psychophysical experiments demonstrating categorical nature of
speech perception.
(2) Hemispheric differences in auditory processing:
• left hemisphere specialized for speech and language
• right hemisphere seems to be more responsive for non-speech
sounds
Categorical Perception
look more closely at physical/acoustic properties of speech signal
sound spectrogram: picture of distribution of acoustic energy
/da/
/ga/
(Hz)
/ba/
(ms)
density or darkness: amplitude (loudness)
Formants: bands of energy - due to resonances of vocal tract
/ba/
/da/
/ga/
f3
f2
f1
rising transition
falling transition
Manipulation: constant series of 13 syllables that differ only
in f2 transition
Task: present to subjects for identification - many presentations
Identification function: rapid change in identifying function
suggests subjects have learned category
Conclusions:
(1) /ba/, /da/, /ga/ (and other phonemes) are perceived categorically,
unlike non-speech sounds
(2) main cue: f2 transition
Speech Perception
Neurons in the auditory cortex are specific for speech sounds.
For example: /ba/ /pa/ /ta/ and /da/.
There is evidence that these neurons compose ‘speech-selective’
channels.
Adaptation!!!!
Pre-adaptation: listener had to identify an ambiguous sound as
either /da/ or /ta/ (50% chance of choosing either one).
Adaptation: listen to /da/ over and over again.
Post-adaptation: listen to ambiguous sound again.
Result: listeners will identify the ambiguous sound as /ta/.
The /da/ channel was fatigued.
The End
 (1) limits of hearing
 (2) loudness
 (3) pitch
 (4) perceptual organization
 (5) localization
 (6) speech perception