DocumentCode :
3583654
Title :
Matched filter design for diphone subspace models
Author :
Reinhard, Klaus ; Niranjan, Mahesan
Author_Institution :
Dept. of Eng., Cambridge Univ., UK
Volume :
6
fYear :
2000
fDate :
6/22/1905 12:00:00 AM
Firstpage :
3430
Abstract :
Considering the perceptual importance of phonetic transitions as minimal contextual variant units, this paper addresses the problem by modelling explicitly interphone dynamics covered in diphones. Subspace projections based on a time-constrained PCA (TC-PCA) are developed which focus on the temporal evolution. They reveal characteristic trajectories present in a low-dimensional spectral representation facilitating robust parameter estimation and simultaneously optimise the discriminant information. A matched filter design is applied to a multiple hypotheses rescoring scheme which enables operating in very low-dimensional parameter space. Using such multiple hypotheses paradigm the complementary information effectiveness of modelling explicitly inter-phone dynamics covered in diphones can be shown using the TIMIT database, resulting in improved phone error rates
Keywords :
filtering theory; matched filters; principal component analysis; signal representation; spectral analysis; speech processing; TC-PCA; TIMIT database; diphone subspace models; discriminant information optimisation; interphone dynamics; low-dimensional parameter space; low-dimensional spectral representation; matched filter design; minimal contextual variant units; multiple hypotheses rescoring; phone error rates; phonetic transitions; principal component analysis; robust parameter estimation; speech signals; subspace projections; temporal evolution; time-constrained PCA; Computer science; Context modeling; Hidden Markov models; Matched filters; Parameter estimation; Principal component analysis; Robustness; Spatial databases; Speech; Steady-state;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Acoustics, Speech, and Signal Processing, 2000. ICASSP '00. Proceedings. 2000 IEEE International Conference on
ISSN :
1520-6149
Print_ISBN :
0-7803-6293-4
Type :
conf
DOI :
10.1109/ICASSP.2000.860138
Filename :
860138
Link To Document :
بازگشت