Abstract
We consider adaptation by topologically smooth transformation with applications to environment and speaker adaptation for robust speech recognition. Specifically, the trade-off between global affine transformations that fail to capture local variation but preserve topology and local class dependent bias transformations that have more resolution but may destroy the topology of the reference model is addressed. We cast the problem of topology preservation of the reference model in an information divergence geometry framework and derive a class of alternating minimization algorithms that aims to preserve topology explicitly during adaptation.
| Original language | English (US) |
|---|---|
| Title of host publication | ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings |
| Publisher | IEEE |
| Pages | 3743-3746 |
| Number of pages | 4 |
| Volume | 6 |
| State | Published - 2000 |
| Externally published | Yes |
| Event | 2000 IEEE Interntional Conference on Acoustics, Speech, and Signal Processing - Istanbul, Turkey Duration: Jun 5 2000 → Jun 9 2000 |
Other
| Other | 2000 IEEE Interntional Conference on Acoustics, Speech, and Signal Processing |
|---|---|
| City | Istanbul, Turkey |
| Period | 6/5/00 → 6/9/00 |
Fingerprint
ASJC Scopus subject areas
- Signal Processing
- Electrical and Electronic Engineering
- Acoustics and Ultrasonics
Cite this
Information geometry of topology preserving adaptation. / Sonmez, Mustafa (Kemal).
ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings. Vol. 6 IEEE, 2000. p. 3743-3746.Research output: Chapter in Book/Report/Conference proceeding › Conference contribution
}
TY - GEN
T1 - Information geometry of topology preserving adaptation
AU - Sonmez, Mustafa (Kemal)
PY - 2000
Y1 - 2000
N2 - We consider adaptation by topologically smooth transformation with applications to environment and speaker adaptation for robust speech recognition. Specifically, the trade-off between global affine transformations that fail to capture local variation but preserve topology and local class dependent bias transformations that have more resolution but may destroy the topology of the reference model is addressed. We cast the problem of topology preservation of the reference model in an information divergence geometry framework and derive a class of alternating minimization algorithms that aims to preserve topology explicitly during adaptation.
AB - We consider adaptation by topologically smooth transformation with applications to environment and speaker adaptation for robust speech recognition. Specifically, the trade-off between global affine transformations that fail to capture local variation but preserve topology and local class dependent bias transformations that have more resolution but may destroy the topology of the reference model is addressed. We cast the problem of topology preservation of the reference model in an information divergence geometry framework and derive a class of alternating minimization algorithms that aims to preserve topology explicitly during adaptation.
UR - http://www.scopus.com/inward/record.url?scp=0033678499&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0033678499&partnerID=8YFLogxK
M3 - Conference contribution
VL - 6
SP - 3743
EP - 3746
BT - ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings
PB - IEEE
ER -