发明名称 |
Method and System for Generating Training Data for an Automatic Speech Recogniser |
摘要 |
The invention describes a method and a system for generating training data (D<SUB>T</SUB>) for an automatic speech recogniser ( 2 ) for operating at a particular first sampling frequency (f<SUB>H</SUB>), comprising steps of deriving spectral characteristics (S<SUB>L</SUB>) from audio data (D<SUB>L</SUB>) sampled at a second frequency (f<SUB>L</SUB>) lower than the first sampling frequency (f<SUB>H</SUB>), extending the bandwidth of the spectral characteristics (S<SUB>L</SUB>) by retrieving bandwidth extending information<SUB>OBE) </SUB>from a codebook (6), and processing the bandwidth extended spectral characteristics (S<SUB>LE</SUB>) to give the required training data (D<SUB>T</SUB>). Moreover a method and a system ( 5 ) for generating a codebook ( 6 ) for extending the bandwidth of spectral characteristics (S<SUB>L</SUB>) for audio data (D<SUB>L</SUB>) sampled at a second sampling frequency (f<SUB>L</SUB>) to spectral characteristics (S<SUB>H</SUB>) for a first sampling frequency (f<SUB>H</SUB>) higher than the second sampling frequency (f<SUB>L</SUB>) are described.
|
申请公布号 |
US2008215322(A1) |
申请公布日期 |
2008.09.04 |
申请号 |
US20080597983 |
申请日期 |
2008.05.16 |
申请人 |
KONINKLIJKE PHILIPS ELECTRONIC, N.V. |
发明人 |
FISCHER ALEXANDER;BIPPUS ROLF DIETER |
分类号 |
G10L15/06;G10L15/065;G10L21/02;G10L21/038 |
主分类号 |
G10L15/06 |
代理机构 |
|
代理人 |
|
主权项 |
|
地址 |
|