Automatic identification of vocalic intervals in speech signal

1
Automatic identification of vocalic intervals in
speech signal
Jesus Garcia Antonio Galves
Phonological Support
Flaviane Fernandes Janaisa Viscardi Ulrike Gut
Many thanks to
Dafydd Gibbon Emmanuel Dupoux Franck Ramus
2
Problem
We have
and we want
0.179 0.301 v 0.301 0.390 c 0.390 0.440
v 0.440 0.498 c 0.498 ....
3
Identification
Cues to the segmentation

• acoustic signal plot
• spectrogram plot
• listening

4
Example
A autoridade do governador diminuiu
Vowel.
Cons.
Vowel
Noise
Acoustic Signal
Spectrogram
5
Vowel.
Vowel
Cons.
Noise
Marca
6
vowels are recurrent
7
t signal
8
c(i,t) energy for the frequency i in time t
9
Relative entropy of the column t respect to the
column s.
10
we use h(t-1,t), h(t-2,t) e h(t-3,t).
h(t)h(t-1,t)h(t-2,t) h(t-3,t)
plot for h(t)
11
(No Transcript)
12
Distance between consecutives columns
13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
Spectrogram for the sentence' A hurricane was
announced this afternoon on the TV.'
20
Signal
Relative Entropy 800-3000 hz
Relative Entropy 0-800 hz
21
Spectrogram and segmentation for' A hurricane
was announced this afternoon on the TV.'
22
(No Transcript)
23
Box Plots for the integral of the relative
entropy for each language
24
Spectrogram and R. entropy for the sentence with
the lower integral
25
Ramus classification (1999).
V vowel percent DeltaC standard
deviation of consonantal intervals
26
(No Transcript)