s4: speech translation and lvcsrsap.ist.i.kyoto-u.ac.jp/asru2007/slide/p4-nakamura.pdf · s4 panel...
TRANSCRIPT
![Page 1: S4: Speech Translation and LVCSRsap.ist.i.kyoto-u.ac.jp/asru2007/slide/P4-Nakamura.pdf · S4 Panel Discussion: “Problems and Future Directions of Speech Translation Technology”](https://reader034.vdocuments.us/reader034/viewer/2022042208/5eabf6eb4b1f731ff221f2b4/html5/thumbnails/1.jpg)
S4: Speech Translation and LVCSR• 1 invited talk by Dr. Stephan Vogel (CMU)
“Speech‐Translation: From Domain‐Limited to Domain‐Unlimited Translation Tasks”
• Accepted papers (20 papers)
– Speech translation: 4 papers
– Language ID: 2 papers
– Cross‐lingual, accented ASR aspects: 4 papers
– Nonnative speech DB: 1 paper
– General LVCSR: 9 papers • Keywords:
WFST, 5 pass decoding, System combination, Arabic, Chinese, Parliamentary speeches
• Panel Discussion:“Problems and future directions of speech translation technology”
![Page 2: S4: Speech Translation and LVCSRsap.ist.i.kyoto-u.ac.jp/asru2007/slide/P4-Nakamura.pdf · S4 Panel Discussion: “Problems and Future Directions of Speech Translation Technology”](https://reader034.vdocuments.us/reader034/viewer/2022042208/5eabf6eb4b1f731ff221f2b4/html5/thumbnails/2.jpg)
S4 Panel Discussion: “Problems and Future Directions of
Speech Translation Technology”
Rohit Prasad (BBN)Chiori Hori (CMU/ATR‐NICT)
Evgeny Matusov (RWTH Aachen)George Saon (IBM)
Stephan Vogel (CMU)
Coordinator: Satoshi Nakamura (ATR‐NICT)
![Page 3: S4: Speech Translation and LVCSRsap.ist.i.kyoto-u.ac.jp/asru2007/slide/P4-Nakamura.pdf · S4 Panel Discussion: “Problems and Future Directions of Speech Translation Technology”](https://reader034.vdocuments.us/reader034/viewer/2022042208/5eabf6eb4b1f731ff221f2b4/html5/thumbnails/3.jpg)
C‐Star III (2000‐)
C‐Star II (1993)
IWSLT (2004)
IWSLT (2005)
IWSLT (2006)
IWSLT (2007)
Spee
ch‐t
o‐Sp
eech
Tran
slat
ion
Spee
ch‐T
rans
latio
n
GALE(2006‐?)
TransTac(2006‐)
ATR Phase I(1986‐1992)
ATR Phase II(1992‐1999)
ATR Phase III(2000.4‐2006.3)
NICT‐ATR Phase IV(2006.4‐)
TC‐STAR (2004‐2006)
Nespole! (2001‐2003)
2005 2007
VerbMobil( 1993‐2000 )
C‐Star I (1991)
1990 2000
![Page 4: S4: Speech Translation and LVCSRsap.ist.i.kyoto-u.ac.jp/asru2007/slide/P4-Nakamura.pdf · S4 Panel Discussion: “Problems and Future Directions of Speech Translation Technology”](https://reader034.vdocuments.us/reader034/viewer/2022042208/5eabf6eb4b1f731ff221f2b4/html5/thumbnails/4.jpg)
IWSLT 2008 Venue(co‐located with AMTA 2008)
IWSLT 2008 Venue(co‐located with AMTA 2008)
Hilton Prince Kuhio Hotel2500 Kuhio Ave,Honolulu, HI 96815
12/25/2007 4
![Page 5: S4: Speech Translation and LVCSRsap.ist.i.kyoto-u.ac.jp/asru2007/slide/P4-Nakamura.pdf · S4 Panel Discussion: “Problems and Future Directions of Speech Translation Technology”](https://reader034.vdocuments.us/reader034/viewer/2022042208/5eabf6eb4b1f731ff221f2b4/html5/thumbnails/5.jpg)
ATR Speech Translation System (2006.12)
Stand aloneStand aloneSmall PCSmall PC
12/25/20075
![Page 6: S4: Speech Translation and LVCSRsap.ist.i.kyoto-u.ac.jp/asru2007/slide/P4-Nakamura.pdf · S4 Panel Discussion: “Problems and Future Directions of Speech Translation Technology”](https://reader034.vdocuments.us/reader034/viewer/2022042208/5eabf6eb4b1f731ff221f2b4/html5/thumbnails/6.jpg)
New Speech Translation System (2007.12)
12/25/2007 6
![Page 7: S4: Speech Translation and LVCSRsap.ist.i.kyoto-u.ac.jp/asru2007/slide/P4-Nakamura.pdf · S4 Panel Discussion: “Problems and Future Directions of Speech Translation Technology”](https://reader034.vdocuments.us/reader034/viewer/2022042208/5eabf6eb4b1f731ff221f2b4/html5/thumbnails/7.jpg)
System Design
Noise Suppression
FeatureExtraction
Compression
Decompression
Cell phonePacket network
+Roaming
DSR in Cell phone client
DecodingMTTTS
Server
NTT Docomo New 905i SeriesWorld Wide Cell Phone (GSM, 3G, JP)
Visit Demo session this evening!
![Page 8: S4: Speech Translation and LVCSRsap.ist.i.kyoto-u.ac.jp/asru2007/slide/P4-Nakamura.pdf · S4 Panel Discussion: “Problems and Future Directions of Speech Translation Technology”](https://reader034.vdocuments.us/reader034/viewer/2022042208/5eabf6eb4b1f731ff221f2b4/html5/thumbnails/8.jpg)
Speech(‐to‐speech) Translation
Multi‐lingual ASR
Multi‐lingual MT
Multi‐lingual TTS
InformationAnalysis
•Performances of modules•Module Integration•Performance Measurement•Real World
![Page 9: S4: Speech Translation and LVCSRsap.ist.i.kyoto-u.ac.jp/asru2007/slide/P4-Nakamura.pdf · S4 Panel Discussion: “Problems and Future Directions of Speech Translation Technology”](https://reader034.vdocuments.us/reader034/viewer/2022042208/5eabf6eb4b1f731ff221f2b4/html5/thumbnails/9.jpg)
Panelists’ presentation
• Rohit Prasad (BBN)
• Chiori Hori (CMU/ATR‐NICT)
• Evgeny Matusov (RWTH Aachen)
• George Saon (IBM)
• Stephan Vogel (CMU)
![Page 10: S4: Speech Translation and LVCSRsap.ist.i.kyoto-u.ac.jp/asru2007/slide/P4-Nakamura.pdf · S4 Panel Discussion: “Problems and Future Directions of Speech Translation Technology”](https://reader034.vdocuments.us/reader034/viewer/2022042208/5eabf6eb4b1f731ff221f2b4/html5/thumbnails/10.jpg)
Challenges for Speech(‐to‐speech) Translation
Integration, Speech translation:Close coupling of ASR and SMT both in training and decoding.ConsolidationMore speech translation, Simultaneous speech translation
Measure: Better objective measure for speech translation Measures considering communicability.
MT: Integration of SMT and rule‐based systems.Help of meaning and structure for SMT.SMT for the very long sentences.Disambiguation utilizing contexts and situations.Difference between human processing and current state‐of‐the art S2S.
Real application:Way to extend more languages.Named entity, OOV problems.Barrier to speech translation business.
Others:Merits and demerits of national projects.