%PDF-1.6
%
1 0 obj
<>
endobj
2 0 obj
<>stream
2021-06-18T13:40:29+09:00
2021-06-18T13:40:29+09:00
2021-06-18T13:40:29+09:00
Adobe Acrobat 21.0
application/pdf
Ritsumeikan Univ.KO-402
uuid:05e6eea1-d62a-4978-a905-a1587d5b8f64
uuid:3dc91588-5796-4b28-9142-320709f22b45
Acrobat Web Capture 15.0
endstream
endobj
5 0 obj
<>
endobj
6 0 obj
<>
endobj
3 0 obj
<>
endobj
7 0 obj
<>
endobj
10 0 obj
<>/ProcSet[/PDF/Text]>>/Type/Page>>
endobj
11 0 obj
<>stream
1 g
10 36 574.5 780 re
f
BT
0 g
/TT0 1 Tf
14.25 0 0 14.25 159.4641 788.25 Tm
(THE SPEECH VISUALIZATION SYSTEM )Tj
/C0_0 1 Tf
6.822 -1.316 Td
<02B5>Tj
/TT0 1 Tf
1 0.105 Td
( KANNON)Tj
/TT1 1 Tf
12 0 0 12 384.0508 742.5 Tm
(Ken Nakamuro)Tj
/C0_0 1 Tf
6.137 -0.187 Td
<02A2>Tj
/TT1 1 Tf
1 0.187 Td
(Kenji Nakamuro)Tj
/C0_0 1 Tf
6.692 -0.187 Td
<02A3>Tj
/TT1 1 Tf
-42.125 -2.125 Td
(This thesis addresses the speech visualization system ``KanNon" which su\
pports communication of the)Tj
-1 -1.5 Td
(deaf people. Especially to the spectral analysis applying the minimum cr\
oss entropy method, its application to)Tj
0 -1.5 TD
(the speech signal, the speech feature extraction and the speech recognit\
ion system based on neural network)Tj
T*
(techniques are presented.)Tj
1 -2.5 Td
(In the First, the concepts and structure of the KanNon system are propos\
ed considering the present)Tj
-1 -1.5 Td
(situation of deaf people in Japan.)Tj
1 -2.5 Td
(Secondly, as the main part of the feature extraction of the speech, cont\
inuous spectral estimation method)Tj
-1 -1.5 Td
(applying minimum cross entropy \(MCE\) analysis considering an auditory \
perceptual property are derived.)Tj
T*
(Then we derived the MCE method with uncertain constrains of autocorrelat\
ion function. And the numerical)Tj
T*
(experiments comparing both methods are performed.)Tj
1 -2.5 Td
(Furthermore, a novel AR model parameter estimation method extending the \
Burg method on the basis of)Tj
-1 -1.5 Td
(the MCE principle is proposed. In order to apply the proposed method to \
a spectral estimation of a speech)Tj
T*
(data, we introduce an algorithm to determine the usage of a prior inform\
ation, based on the divergence)Tj
T*
(measure defined by the Kullback information, since effectiveness of a pr\
ior information to spectral estimation)Tj
T*
(results depends on the variation of speech signal. The estimation result\
s for real speech data illustrate)Tj
T*
(improved performance in comparison to the Burg method.)Tj
1 -2.5 Td
(On the one hand, pitch frequency estimation method considering continuit\
y of the pitch frequency is)Tj
-1 -1.5 Td
(addressed and experimental results are presented.)Tj
1 -2.5 Td
(Finally, the speech recognition system consists of speech/silence, voice\
/unvoiced, vowel recognition based)Tj
-1 -1.5 Td
(on the neural network techniques is proposed. And the experimental resul\
ts of recognition test using real)Tj
T*
(speech data are shown.)Tj
ET
endstream
endobj
12 0 obj
<>
endobj
13 0 obj
($-t}5>{O)
endobj
14 0 obj
<>
endobj
15 0 obj
<>
endobj
16 0 obj
<>
endobj
21 0 obj
<>
endobj
22 0 obj
<>stream
H\j0E
-Eq3)@kK~cO/QEjH|bzS[;ucG˪YzN2Z:fUgtK\ ?kD~?S'NcOO,dG/"Gád}K-ѷdX]],p
jZaLdg(%%%%%%<[
3
2
3
2
3
2*J_W+J_W+J?0`_QK_c鳙fS~p
!
endstream
endobj
23 0 obj
<>stream
H|y\TG5o^p rfFzFYbPkv5xD%*(1ފ
^㉷x((3:F̤GYS]U^}? !sM|9BXlB$E'8aѯ E A+bF
@_z|r u ='@\`qѱO#!SX[6 )eT+ I␘h}n5U[$EJ
l&G'u;R }!S\au'Kd|AQd%rSv'YaII{
\G18CvjkD$2y=5ՙ@$r!HQM!
5o)ie=<}U\FZ
!d
SW֠aM~Ь-Z~[}ٺMO۵G>t_٫W}1qL4xHaS1r1c7>5mw'}?yig̜5;cyd.\x -_2+{5kYoظIڼ%g;vڽ'w:|1;ŒK\V[@E:c<4FM)T,Ri)[*^i}r_ye/#pit&;G8ܜ\w3!W9kŬDOx<e@YqY/3陸TVv_e{uvvɮ'2K3KWNJ=KZҦav[K[s[S[#[]Vh#Szzǽ˚o=b=lzҺֺf5[M{k\ "J0#hF+LTD{Ew~A0&