lecture 18 - university of california, berkeley · consonants are more important than vowels. ......
TRANSCRIPT
EE22 LECTURE ON SPEECH PERCEPTION
N.M 18.1
PE Spring,1999
5D
ORGAN / B.GOLD LECTURE 18
University of CaliforniaBerkeley
College of EngineeringDepartment of Electrical Engineering
and Computer Sciences
rofessors : N.Morgan / B.GoldE225D
Speech Perception
Lecture 18
EE2 LECTURE ON SPEECH PERCEPTION
N.M 18.2
in the auditory system.
25D
ORGAN / B.GOLD LECTURE 18
Figure 17.1 : Block diagram of sound representation
EE2 LECTURE ON SPEECH PERCEPTION
N.M 18.3
25D
ORGAN / B.GOLD LECTURE 18
Speech Perception
Production
Physiology Psychophysics
Consonant Perception
Consonants are more important than vowels.
Cnsnnts r mr mprtnt thn vwls
ooa ae oe ioa a oe
Perception
EE2 LECTURE ON SPEECH PERCEPTION
N.M 18.4
uency response
25D
ORGAN / B.GOLD LECTURE 18
Figure 17.5 : Confusion matrix for S/N=+12dB and freq
of 200-6500Hz.
EE2 LECTURE ON SPEECH PERCEPTION
N.M 18.5
ency response
25D
ORGAN / B.GOLD LECTURE 18
Figure 17.5 : Confusion matrix for S/N=-6dB and frequof 200-6500Hz.
EE2 LECTURE ON SPEECH PERCEPTION
N.M 18.6
tering.
25D
ORGAN / B.GOLD LECTURE 18
Condition S/N Band 1 -18 200 - 6500 2 -12 200 - 6500 3 - 6 200 - 6500 4 0 200 - 6500 5 6 200 - 6500 6 12 200 - 6500
7 12 200 - 300 8 12 200 - 400 9 12 200 - 60010 12 200 - 120011 12 200 - 2500
12 12 200 - 5000
13 12 1000 - 500014 12 2000 - 500015 12 2500 - 500016 12 3000 - 500017 12 4000 - 5000
Figure 17.6 : Seventeen conditions of S/N and fil
EE2 LECTURE ON SPEECH PERCEPTION
N.M 18.7
Sounds
25D
ORGAN / B.GOLD LECTURE 18
Miller Nicely System
Voicing
Nasality
Affrication
Duration
Place of Articulation
s - z
m, n, ng, other
fricatives m,n
p, b
p, k, t
b, g, d
EE2 LECTURE ON SPEECH PERCEPTION
N.M 18.8
25D
ORGAN / B.GOLD LECTURE 18
EE2 LECTURE ON SPEECH PERCEPTION
N.M 18.9
25D
ORGAN / B.GOLD LECTURE 18
EE2 LECTURE ON SPEECH PERCEPTION
N.M 18.10
25D
ORGAN / B.GOLD LECTURE 18
EE2 LECTURE ON SPEECH PERCEPTION
N.M 18.11
bson et al.
25D
ORGAN / B.GOLD LECTURE 18
Figure 17.7 : Binary distinctive feature set of Jako
EE2 LECTURE ON SPEECH PERCEPTION
N.M 18.12
plosure - fs vowel
F3
F2
F1
25D
ORGAN / B.GOLD LECTURE 18
Jacobson, Fant, Halle.
Production Perception
Haskin’s Lab. Pattern Playback
spectrum
burst
(freq.)
formant transition
t
k
p
silence
EE2 LECTURE ON SPEECH PERCEPTION
N.M 18.13
urst frequencies s.
25D
ORGAN / B.GOLD LECTURE 18
Figure 17.9 : Perceptual responses to different band vowels for the voiceless plosive
EE2 LECTURE ON SPEECH PERCEPTION
N.M 18.14
SR’s for the fine 20ms. ulus.
.)
ALSR
25D
ORGAN / B.GOLD LECTURE 18
Figure 17.11 : Smoothed spectra and corresponding ALIntervals at the begining o f the - da - stim
Freq. (kHzFreq. (kHz.)
Smoothed Stimulus Smoothed
EE22 LECTURE ON SPEECH PERCEPTION
N.M 18.15
ech perception.
SPrevious
ller / Phone
Phonemes
logical
Output
Context
n Maker
of the uage
5D
ORGAN / B.GOLD LECTURE 18
Figure 17.12 : Analysis by synthesis motor theory of spe
peech Analysis ofStore
Tentative
Tentative
Comparator
Differences Contro Trail Pattern of
Trial Symthesis of
Phono
Articulatory
Input Auditory Features
Decoding to Phonemes
Phonemes
Decisio
Auditory Pattern from Motor Instructions
Motor Instructions
Rules Lang
Auditory Features
A B C
D
E
F
G
H
EE22 LECTURE ON SPEECH PERCEPTION
N.M 18.16
: Neural firing patterns
F’s and spectrogram
nce.
5D
ORGAN / B.GOLD LECTURE 18
.
Figure 17.13
for different C
of same sente