An N-Best Strategy, Dynamic Grammars and Selectively Trained Neural Networks for Real-Time Recognition of Continuously Spelled Names over the Telephone
Identifieur interne : 00C327 ( Main/Exploration ); précédent : 00C326; suivant : 00C328An N-Best Strategy, Dynamic Grammars and Selectively Trained Neural Networks for Real-Time Recognition of Continuously Spelled Names over the Telephone
Auteurs : J.-C. Junqua ; S. Valente ; D. Fohr ; J.-F. MariSource :
Abstract
We introduce SmarTspelL, a new speaker-independent algorithm to recognize continuously spelled names over the telephone. Our method is based on an N-best multi-pass recognition strategy applying costly constraints when the number of possible candidates is low. This strategy outperforms an HMM recognizer using a grammar containing all the possible names. It is also more suitable to real-time. For a 3,388 name dictionary, a 95.3% name recognition rate is obtained. A real-time prototype has been implemented on a workstation. We also present comparisons of different feature sets for speech representation, and two speech recognition approaches based on first- and second- order HMMs.
Affiliations:
Links toward previous steps (curation, corpus...)
- to stream Crin, to step Corpus: 001A28
- to stream Crin, to step Curation: 001A28
- to stream Crin, to step Checkpoint: 002B39
- to stream Main, to step Merge: 00CB84
- to stream Main, to step Curation: 00C327
Le document en format XML
<record><TEI><teiHeader><fileDesc><titleStmt><title xml:lang="en" wicri:score="518">An N-Best Strategy, Dynamic Grammars and Selectively Trained Neural Networks for Real-Time Recognition of Continuously Spelled Names over the Telephone</title>
</titleStmt>
<publicationStmt><idno type="RBID">CRIN:junqua95b</idno>
<date when="1995" year="1995">1995</date>
<idno type="wicri:Area/Crin/Corpus">001A28</idno>
<idno type="wicri:Area/Crin/Curation">001A28</idno>
<idno type="wicri:explorRef" wicri:stream="Crin" wicri:step="Curation">001A28</idno>
<idno type="wicri:Area/Crin/Checkpoint">002B39</idno>
<idno type="wicri:explorRef" wicri:stream="Crin" wicri:step="Checkpoint">002B39</idno>
<idno type="wicri:Area/Main/Merge">00CB84</idno>
<idno type="wicri:Area/Main/Curation">00C327</idno>
<idno type="wicri:Area/Main/Exploration">00C327</idno>
</publicationStmt>
<sourceDesc><biblStruct><analytic><title xml:lang="en">An N-Best Strategy, Dynamic Grammars and Selectively Trained Neural Networks for Real-Time Recognition of Continuously Spelled Names over the Telephone</title>
<author><name sortKey="Junqua, J C" sort="Junqua, J C" uniqKey="Junqua J" first="J.-C." last="Junqua">J.-C. Junqua</name>
</author>
<author><name sortKey="Valente, S" sort="Valente, S" uniqKey="Valente S" first="S." last="Valente">S. Valente</name>
</author>
<author><name sortKey="Fohr, D" sort="Fohr, D" uniqKey="Fohr D" first="D." last="Fohr">D. Fohr</name>
</author>
<author><name sortKey="Mari, J F" sort="Mari, J F" uniqKey="Mari J" first="J.-F." last="Mari">J.-F. Mari</name>
</author>
</analytic>
</biblStruct>
</sourceDesc>
</fileDesc>
<profileDesc><textClass></textClass>
</profileDesc>
</teiHeader>
<front><div type="abstract" xml:lang="en" wicri:score="2114">We introduce SmarTspelL, a new speaker-independent algorithm to recognize continuously spelled names over the telephone. Our method is based on an N-best multi-pass recognition strategy applying costly constraints when the number of possible candidates is low. This strategy outperforms an HMM recognizer using a grammar containing all the possible names. It is also more suitable to real-time. For a 3,388 name dictionary, a 95.3% name recognition rate is obtained. A real-time prototype has been implemented on a workstation. We also present comparisons of different feature sets for speech representation, and two speech recognition approaches based on first- and second- order HMMs.</div>
</front>
</TEI>
<affiliations><list></list>
<tree><noCountry><name sortKey="Fohr, D" sort="Fohr, D" uniqKey="Fohr D" first="D." last="Fohr">D. Fohr</name>
<name sortKey="Junqua, J C" sort="Junqua, J C" uniqKey="Junqua J" first="J.-C." last="Junqua">J.-C. Junqua</name>
<name sortKey="Mari, J F" sort="Mari, J F" uniqKey="Mari J" first="J.-F." last="Mari">J.-F. Mari</name>
<name sortKey="Valente, S" sort="Valente, S" uniqKey="Valente S" first="S." last="Valente">S. Valente</name>
</noCountry>
</tree>
</affiliations>
</record>
Pour manipuler ce document sous Unix (Dilib)
EXPLOR_STEP=$WICRI_ROOT/Wicri/Lorraine/explor/InforLorV4/Data/Main/Exploration
HfdSelect -h $EXPLOR_STEP/biblio.hfd -nk 00C327 | SxmlIndent | more
Ou
HfdSelect -h $EXPLOR_AREA/Data/Main/Exploration/biblio.hfd -nk 00C327 | SxmlIndent | more
Pour mettre un lien sur cette page dans le réseau Wicri
{{Explor lien |wiki= Wicri/Lorraine |area= InforLorV4 |flux= Main |étape= Exploration |type= RBID |clé= CRIN:junqua95b |texte= An N-Best Strategy, Dynamic Grammars and Selectively Trained Neural Networks for Real-Time Recognition of Continuously Spelled Names over the Telephone }}
This area was generated with Dilib version V0.6.33. |