Psychoacoustics of Dynamic

1
Psychoacoustics of Dynamic Center-of-Gravity
Signals

• Larry Feth
• Ashok Krishnamurthy
• Ohio State University

2
Spectral Center-of-Gravity

• Chistovitch and Lublinskaja (1976,1979)
• Perceptual Formant at Center-of-Gravity
• Two-formant synthetic vowel
• Matched by adjustable single-formant signal
• Center frequency of match depends on relative
  amplitudes of the two formants

3
Experimental Paradigm
4
Chistovitch and Lublinskaja Results
5
Voelcker Two-tone Signals
6
Voelcker Two-tone Signals

• Initially, led to the EWAIF model
• Envelope-Weighted Average of Instantaneous
  Frequency (time domain)
• Point by point multiply E x F values
• Sum over N periods
• Divide by sum of weights
• Indicates pitch change in periodic signals
• Helmholtz (1954, 2nd English edition)
• Jeffress (1964)

7
EWAIF Model
8
IWAIF Model Predictions
9
Two-tone resolution task

• Feth and OMalley (1977)
• Two-tone resolution
• DI 1 dB Df independent variable
• Voelcker-tone pair pitch discrimination
• inverted u-shaped psychometric functions
• Components resolved beyond 75 point
• 3.5 Bark separation jnnd

10
Voelcker Signal Discrimination Task
11
Discrimination Results

• Jnnd Just not noticeable difference
• Filled circles
• Breakpoint estimates
• Open circles
• CR critical ratio CBW
• CB empirical CBW
• Solid line TW envelope

12
IWAIF Model

• Intensity Weighted Average of Instantaneous
  Frequency Centroid of signals positive power
  spectrum (Anantharaman, et al., 1993)

13
Dynamic Center-of-Gravity Effect

• Lublinskaja (1996)
• Three-formant synthetic Russian vowels
• Listeners identified vowels with
• conventional formant transitions
• co-modulated formant pairs that exhibit the same
  dynamic spectral center-of-gravity
• ID functions were very similar with formant pairs
  separated by 4.3 Bark or less

14
Psychophysics

• Anantharaman (1998)
• Two-tone signals with dynamic c-o-g effect
• We called them Virtual Frequency Glides
• Listeners matched transition rates in VF glides
  to those in FM glides
• IWAIF model predicts results for transitions from
  2 to 5 ERB

15
Dynamic Center-of-Gravity Signals
Waveform
Long-term Spectrum
Spectrogram
16
Rate-matching results
17
Model Results
18
Short-term running IWAIF Model
19
IWAIF Model Results
20
Application of ST-IWAIF Model
21
More Psychophysics

• Research Question(s)
• What is being integrated in spectral
  integration?
• OR
• Where in the auditory system is the processing
  located?

22
Psychophysics

• Iyer, et al., (2001)
• Temporal acuity for FM and VF glides
• Step vs. linear ramp discrimination
• Similar DT values may mean common process
• Masking patterns for FM and VF glides
• Peripheral process i.e., Energy Masking
• Different results VF not peripheral process

23
Temporal Acuity Paradigm
Step (red) versus Glide (blue) transitions for FM
tone (left panel) and Virtual Frequency (right
panel)
24
Temporal Acuity Results

• Just discriminable step duration for FM (solid
  lines filled symbols) and VF (dashed lines
  unfilled symbols) signals. Frequency separations
  are 2, 5 and 8 ERBu. The results for 1000 Hz are
  represented by circles and those for 4000 Hz by
  triangles. Average for 4 listeners.

25
Dynamic Center-of-Gravity Maskers

• Masking of brief probe by FM glide (left
  panel) and by VF glide (right panel). Probe is
  in the spectro-temporal center of each masker.
  Five auditory filter bands are illustrated.

26
Masking Results
27
Glide Direction Asymmetry

• Gordon and Poeppel
• 3 Frequency ranges (for F1,F2 F3)
• 30 unpracticed listeners 20 trials / signal
• One interval Direction Identification Up vs. Dn
• Best results at high frequency (F3) range
• 10- through 160 ms Up is easier to ID than Dn
• Less clear-cut results at low or mid-freq. ranges

28
Glide Direction Asymmetry

• Gordon and Poeppel ARLO (2002)
• Identification of FM Sweep direction is easier
  for rising than for falling tones.

29
Glide Direction Asymmetry

• Dawson, (2002)
• Tested only high frequency range (F3)
• Practiced listeners 100 all conditions!
• Modified procedure
• Rove each frequency sweep over 1 octave
• Practice to asymptote

30
Glide ID Results

• Average for 4 listeners
• One-interval ID task
• 250 trials / datum point
• Well-practiced Subjs
• Starting frequency roved over 1-octave range
• Summary
• FM easier than VF
• Up easier than Down

31
CV Identification Experiment

• da ga continuum varying F3 transition
• Duration 50 ms transition into 200 ms base
• F3 onset 2018 to 2658 Hz in 80 Hz steps
• F3 base 2527 Hz (constant)
• Formant transition type
• Klatt synthesizer
• Frequency Modulated tone glide
• Virtual Frequency glide

32
CV Identification Stimuli
Spectrogram 1. Step 1 of Klatt Monaural
Continuum/ga/ endpoint
33
CV Identification Stimuli
Spectrogram 2. Step 1 of FM Monaural
Continuum/ga/ endpoint
34
CV Identification Stimuli
Spectrogram 3. Step 1 of VF Monaural
Continuum/ga/ endpoint
35
CV Identification Stimuli
Spectrogram 4. Step 1 of Dichotic FM
Continuum/ga/ endpoint
36
CV Identification Stimuli
Spectrogram 5. Step 1 of Dichotic VF
Continuum/ga/ endpoint
37
CV Identification Experiment

• Listeners 8 adults with normal hearing
• Procedure One interval, 2-AFC
• 3 transition types Klatt, FM or VF
• 6 of 8 tokens tested
• 20 repetitions / token
• Results are averaged for the 8 listeners

38
CV Identification Results
39
CV Identification Results
40
Psychoacoustics of Dynamic Center-of-Gravity
Signals

• Conclusions
• Excitation is integrated not signal energy
• The processing is central not peripheral
• Masking Patterns are very different
• Temporal Acuity results are similar for FM VF
  glides
• Direction ID Asymmetry is similar for FM VF
  glides

41
Psychoacoustics of Dynamic Center-of-Gravity
Signals

• Conclusions
• CV identification functions are similar for
• Klatt synthesized sounds
• FM formant sounds
• VF formant sounds
• Thus, it doesnt matter how excitation is moved
  from A to B, the brain will interpret it as the
  same sound.
• The effect is evident under dichotic listening
  further support for central processing.

42
Collaborators
43
Psychoacoustics of Dynamic Center-of-Gravity
Signals

• Thank You
• Questions?

44
Up vs. Down FM Glide
45
Up vs. Down FM Glide
46
Up vs. Down VF Glide
47
Up vs. Down VF Glide
48
Effect of Masker Direction

• Masking produced by VF (above) and FM (below)
  maskers with D F 5 ERB. Purple bars are up
  glides yellow bars are down glides. Centered
  probe.

49
Effect of Masker Position

• Masking produced by VF (above) and FM
  (below) maskers with D F 5 ERB. Purple bars
  are up glides yellow bars are down glides.

50
Klatt FM Parameters
51
Virtual Frequency Parameters