Document Sample
Ikalanga Powered By Docstoc
					Basic Audition and Psychoacoustics
As we shift our focus from the acoustic characteristics of speech sounds
to the linguistic percepts that the acoustic input evokes in the listeners,
we first need to consider the physical and psychophysical properties of
the human auditory system.

The capabilities of this system are remarkable. For example, we can
hear a sound which causes eardrum to move only one-tenth the
diameter of a hydrogen molecule.
                     AUDITORY PHYSIOLOGY

The outer ear consists of the pinna and the auditory canal (external
auditory meatus). The auditory canal is a resonator, and can be
compared to a uniform tube closed at one end and open at the other.
Note: to calculate the resonant frequencies of the auditory canal, we
would use the same formula that we used for calculating the resonant
frequencies of the vocal tract in a neutral configuration. The auditory
canal is about 2.5 cm (~1 inch) long, giving us:
   c * (2n - 1) = 34,400 / 4 x 2.5 = ~ 3440 Hz
The pressure at the eardrum for frequencies in this region can be 2-4 x
greater than at the canal entrance.

To think about:
What consequences might this have (if any) for sound systems?

The auditory canal ends at the eardrum or tympanic membrane. The
sound pressure at the drum displaces the drum, which in turn causes
displacement of the bones (ossicles) of the middle ear: the malleus,
incus, and stapes. These bones are suspended by ligaments in the air-
filled middle ear cavity. The mechanical vibrations of the auditory
ossicles are transmitted to the oval window, a membrane that covers the
opening to the inner-ear cochlea.

Worth noting: The fluid-filled cochlea is more resistant than air to
movement so, all else being equal, most of the sound energy transmitted
to the oval window would be reflected back. But all else isn’t equal:
the oval window is smaller than the eardrum; since pressure =
force/area, this size difference results in increased pressure at the
window. Also, the incus behaves like a fulcrum to the (smaller) stapes,
increasing pressure at the oval window.

The inner ear is a system of cavities in the bones of the skull which
influence balance as well as hearing. The cavity that houses the sensory
receptor for hearing is the cochlea, where the mechanical vibrations of
the middle ear and oval window are transformed into nerve impulses.

The cochlea is a fluid-filled coiled cavity. The vibration of the endplate
of the stapes against the oval window results in pressure waves in the
cochlear fluid, which in turn set the cochlear duct --and the basilar
membrane within the duct --into vibration.
     The uncoiled cochlea:
                              scala vestibuli
basal end                                                                apical end
                              scala tympani
                                                   Organ of

The basilar membrane is narrow and stiff at the basal (oval window) end, where it
responds with greatest amplitude to high frequencies. At the apical end, where it
is thicker and less stiff, the greatest amplitude of response is to low frequencies.
Thus the basilar membrane is a spectrum analyzer, performing a kind
of Fourier analysis on input complex waves albeit with limited power
of resolution.

The basilar membrane’s response to input frequencies is non-linear: a
larger portion of the basilar membrane responds to sounds in the 0-1000
Hz range than, for example, in the 10000-11000 Hz range. This non-
linearity has psychoacoustic consequences: human listeners are more
sensitive to differences in the lower than in the higher frequencies.
(We’ll return to this point shortly.)

Like the digital filters we discussed last time, the cochlea’s response has
a time-frequency trade-off: the basal end (which responds to high
frequencies) provides poorer frequency but better time resolution while
the apical end (which responds to low frequencies) does the opposite.
To think about: How does this time-frequency trade-off relate to what
we know about speech sounds? Which speech sounds have primarily
high frequency energy? Low frequencies? Which have rapidly
changing acoustic characteristics?
The mechanical vibration of the basilar membrane is converted into
signals that are transmitted to the auditory cortex in the Organ of Corti,
which lies along the length of the basilar membrane. The Organ of Corti
consists of rows of thousands of hair cells which are innervated by
~30000 nerve fibers. Movement of the basilar membrane causes
deflection of the endings of the hair cells through contact with the

tectorial membrane, which excites the nerve fibers in that region.
                     BASIC PSYCHOACOUSTICS

Psychoacoustics is concerned with the psychological (subjective)
correlates of the physical parameters of acoustics. So we turn now to
consideration of the sensations that sounds evoke in the listener.

While we are centrally interested in the sensations that the complex
waves of speech evoke, much of the work on psychoacoustics has been
based on non-speech stimuli such as pure tones.
Absolute threshold: the minimum detectable level of a sound under
ideal conditions.

Absolute threshold for pure tones:
Humans are most sensitive to sounds in the 1-5 kHz range; i.e., in this
range we hear sounds that have a relatively low sound pressure level
(SPL). Particular sensitivity at 3-4 kHz is due to the lowest frequency
resonance of the ear canal (which we calculated above at 3440 Hz).

Below 1 kHz the threshold increases gradually (i.e., intensity must be
increased for listeners to detect the sound) and above ~10 kHz the
threshold increases sharply.
The auditory threshold curve was determined on the basis of relatively
long tones (500 ms or longer).

At shorter durations, the threshold increases, but in a frequency-
dependent way. For example, at lower frequencies, the threshold
increases at ~ 400 ms but in the frequency region of greatest sensitivity it
doesn’t begin to increase until ~ 200 ms (Stevens, 1998, Acoustic

The sensation of loudness correlates closely with the intensity of a
From the most to the least intense sound that the human ear can hear
(without damage!), the ratio of intensities is 1,000,000,000,000:1
(B.C.J. Moore, An Introduction to the Psychology of Hearing).
But the subjective loudness differences that sounds evoke in listeners is
nowhere near that great.
Therefore, a commonly used scale for measuring intensity, the decibel
scale (dB), reflects the nonlinearity of loudness perception.
Decibel Scale
Decibel:   unit of measurement of relative intensity of a sound,
           compared to an arbitrary reference point
Intensity: magnitude of sound expressed in pressure or power.
Pressure: force per area (unit = dynes/cm2)
Power:     rate at which energy expended (unit = watts/cm2);
           proportional to the square of the pressure
Reference point:
auditory threshold @ 1 kHz = 0 dB = 0.0002 dynes/cm2 (pressure)
                                  = 10-16 watts/cm2 (power)
Decibel formula for pressure:
        dB SPL = 20 ( log10          )   where Po = output intensity
                                         and Pr = reference intensity
For comparison…

Ratio        Log   dB (20 x log of ratio)

1000:1       3     60 dB SPL       e.g., conversation

 100:1       2     40 dB SPL       e.g., quiet office

  10:1       1     20 dB SPL       e.g., faint whisper

   1:1       0      0 dB SPL       absolute threshold
Equal and relative loudness scales:

The decibel scale is not sensitive to the effects of frequency on the
sensation of loudness. For example, a 300 Hz and a 3000 Hz tone at 50
dB differ substantially in loudness.

Phon: The phon scale is determined by having listeners adjust the
intensity of a 1000 Hz tone until it has the same loudness as a
comparison tone of a different frequency. Sounds judged to have equal
loudness in this way are assigned the same “phon” value (e.g., all tones
judged as having the same loudness as a 20 dB 1000 Hz tone have a
loudness of 20 phons).

Sone: The sone scale is determined by having listeners adjust the
loudness of a tone until it is twice as loud, or half as loud, as another
tone. 1 sone = loudness of a 40 dB 1000 Hz tone. 2 sones = sound
judged to be 2x as loud as this.

Just as the relation between loudness and intensity is non-linear, so is
the relation between pitch and frequency.

Mel: The mel scale is based on experiments with pure tones in which
listeners adjust the frequency of a test tone to be half as high (or twice
as high) as that of a comparison tone.
          1000 mel = pitch of 1000 Hz tone
          500 mel      = pitch of tone that sounds half as high
          2000 mel = pitch of tone that sounds twice as high

The mel scale corresponds closely to the Hz scale up to ~500 Hz. At
higher frequencies, the mel scale is more nearly logarithmic.
Bark: The Bark scale is a critical band scale. In considering critical
bands, we turn first to the principles of frequency selectivity and
masking (see B.C.J. Moore’s An Introduction to the Psychology of Hearing
for an excellent overview).

Frequency selectivity is the human auditory system’s ability to resolve
the components of complex sounds.

Masking refers to the conditions under which one sound is made
inaudible or masked by the presence of another sound (the masker). Put
another way, under masking, the audibility threshold of one sound is
raised by the presence of another sound (the masker).

Because signals are most easily masked by sounds whose frequencies
are the same as, or close to, those of the signal, masking reflects the
limits of frequency selectivity: masking occurs when the selectivity of
the auditory system fails to separate signal and masker.
Fletcher (1940) presented listeners with a pure tone signal plus a noise
masker whose bandwidth (BW) was varied. (The noise was centered at
the frequency of the pure tone.) He found that as the noise BW
increased, so did the signal threshold: that is, a more intense signal was
needed for it to be detectable through the noise.
So far no surprise, but what’s interesting is that this finding held only
up to a certain point: at some noise BW, the threshold function flattens
off and further increases in noise BW do not affect the signal threshold.
     Signal Threshold (dB) 

                                                         In this experimental
                                                         paradigm, the critical band is
                                                         the BW at which the signal
                                                         threshold flattens out.

                               Masker Bandwidth (Hz) 
Over the past 60+ years, numerous other experiments have provided
further evidence of critical bands. The current view (e.g., Syrdal & Gopal,
1986, Journal of the Acoustical Society of America 79) is that the auditory system
is composed of a series of bandpass filters with overlapping
bandwidths. A critical band, then, is the bandwidth of each internal
filter. As the center frequencies of the critical bands increase, so do
their bandwidths.

Physiologically, each critical band corresponds to a distance on the
basilar membrane (1.3 mm according to Syrdal and Gopal). This is in
keeping with what we’ve already seen: recall that a larger portion of
the basilar membrane responds to low frequencies than to higher
frequencies, and human listeners are more sensitive to differences in the
lower than in the higher frequencies (see p. 52 of Johnson’s book).
Zwicker (1961) divided the human auditory range below 16 kHz into
24 critical bands or Barks. One of the more widely used frequency-to-
Bark approximations is that of Zwicker & Terhardt (1980) (given in
Syrdal & Gopal, 1986, JASA vol. 79).

The bark scale, like the mel scale, is essentially linear up to about 500
Hz and is nearly logarithmic above 500 Hz.

How small of a difference in frequency, intensity, or duration can the
human auditory system detect? The answer depends on the type of input
signal. In general, JNDs are larger for speech or speech-like stimuli than
for pure tones or, in some cases, noise. Some examples from K.N.
Stevens’ Acoustic Phonetics (1998) are:

Loudness JNDS:
Pure tones or wideband noise at amplitudes in the speech range: 0.3-1.0 dB
Amplitude of F2 in a vowel with equally spaced formants:          ~ 3 dB

Pitch JNDs:
Pure tones (normal listening levels):
   ~ 1 Hz for frequencies up to 1 kHz;
   ~ 2 Hz at about 2 kHz and ~ 4 Hz at about 4 k Hz
   Increases rapidly above 5 kHz
F2: 20-100 Hz, depending on the F1-F2 or F2-F3 distance

Shared By: