发明名称 |
DISCRIMINATIVE TRAINING OF DOCUMENT TRANSCRIPTION SYSTEM |
摘要 |
<p>A system is provided for training an acoustic model (330) for use in speech recognition. In particular, such a system may be used to perform training (328) based on a spoken audio stream (302) and a non-literal transcript (304) of the spoken audio stream (302). Such a system may identify (204) text (308) in the non-literal transcript (304) which represents concepts having multiple spoken forms. The system may attempt to identify the actual spoken form in the audio stream (302) which produced the corresponding text in the non-literal transcript (304), and thereby produce a revised transcript (326) which more accurately represents the spoken audio stream (302). The revised, and more accurate, transcript (326) may be used to train (328) the acoustic model (330) using discriminative training techniques (1414), thereby producing a better acoustic model than that which would be produced using conventional techniques, which perform training based directly on the original non-literal transcript (304).</p> |
申请公布号 |
WO2006034152(A2) |
申请公布日期 |
2006.03.30 |
申请号 |
WO2005US33403 |
申请日期 |
2005.09.16 |
申请人 |
MULTIMODAL TECHNOLOGIES, INC.;YEGNANARAYANAN, GIRIJA;FRITSCH, JUERGEN;MATHIAS, LAMBERT |
发明人 |
YEGNANARAYANAN, GIRIJA;FRITSCH, JUERGEN;MATHIAS, LAMBERT |
分类号 |
G10L15/26;G10L15/06;G10L15/18;G10L15/183 |
主分类号 |
G10L15/26 |
代理机构 |
|
代理人 |
|
主权项 |
|
地址 |
|