SuperLectures.com

FRONT-END FEATURE TRANSFORMS WITH CONTEXT FILTERING FOR SPEAKER ADAPTATION

Adaptation for ASR

Full Paper at IEEE Xplore

Přednášející: Steven Rennie, Autoři: Jing Huang, IBM T.J. Watson Research Center, United States; Karthik Visweswariah, IBM India Research, India; Peder Olsen, Vaibhava Goel, IBM T.J. Watson Research Center, United States

Feature-space transforms such as feature-space maximum likelihood linear regression (FMLLR) are very effective speaker adaptation technique, especially on mismatched test data. In this study, we extend the full-rank square matrix of FMLLR to a non-square matrix that use neighboring feature vectors in estimating the adapted central feature vector. Through optimizing an appropriate objective function we aim to filter out and transform features through the correlation of the feature context. We compare to FMLLR that just consider the current feature vector only. Our experiments are conducted on the automobile data with different speed conditions. Results show that context filtering improves 23% on word error rate over conventional FMLLR on noisy 60mph data with adapted ML model, and 7%/9% improvement over the discriminatively trained FMMI/BMMI models.


  Přepis řeči

|

  Slajdy

Zvětšit slajd | Zobrazit všechny slajdy

0:00:16

  1. slajd

0:01:05

  2. slajd

0:02:08

  3. slajd

0:02:42

  4. slajd

0:06:05

  5. slajd

0:07:28

  6. slajd

0:08:22

  7. slajd

0:08:48

  8. slajd

0:09:35

  9. slajd

0:10:07

 10. slajd

0:10:42

 11. slajd

0:11:28

 12. slajd

0:11:44

 13. slajd

0:12:29

 14. slajd

0:12:48

 15. slajd

0:14:24

    11. slajd

0:16:31

    13. slajd

  Komentáře

Please sign in to post your comment!

  Informace o přednášce

Nahráno: 2011-05-24 16:15 - 16:35, Panorama
Přidáno: 15. 6. 2011 14:44
Počet zhlédnutí: 78
Rozlišení videa: 1024x576 px, 512x288 px
Délka videa: 0:17:10
Audio stopa: MP3 [5.79 MB], 0:17:10