TY  - GEN
AB  - Temporal patterns of frequency-localized features in ASR. This work  investigates the use  of  frequency-localized temporal patterns of  the speech signal  for  developing  robust  front-end  for  Automatic  Speech  Recognition  (ASR). Various linear transforms are investigated  for  parameterization of  the frequency-localized temporal  patterns. We show  that  temporal patterns  closely  follow  the  properties  of a  first- order Markov  process,  which  results in  the PCA transforms being very close to the DCT transform.  Better  recognition performance is achieved on  using  the DCT components of temporal patterns as opposed to directly using temporal patterns  for  feature estimation. Other  linear  transforms  such  as Linear  Discriminant  Analysis  (LDA)  are  also  studied by the parameterization.  The parameterized TempoRAl  Patterns (TRAPS) are used  to estimate broad-phonetic class-posteriors independently in  each critical-band.  These class- posteriors are  combined and  used  as the features for  word  recognition.  Our work  shows that broad-phonetic features generalize better than other conventional features and yield considerable complementary information concerning short-term central features in ASR.
AD  - Oregon Health and Science University
AU  - Jain, Pratibha
DA  - 2003-07-01
DO  - 10.6083/M4GX48VH
DO  - DOI
ED  - Hermansky, Hynek
ED  - Advisor
ID  - 2416
KW  - Speech
KW  - Phonetics
KW  - Markov Chains
KW  - Discriminant Analysis
KW  - Speech Perception
L1  - https://digitalcollections.ohsu.edu/record/2416/files/3154_etd.pdf
L2  - https://digitalcollections.ohsu.edu/record/2416/files/3154_etd.pdf
L4  - https://digitalcollections.ohsu.edu/record/2416/files/3154_etd.pdf
LK  - https://digitalcollections.ohsu.edu/record/2416/files/3154_etd.pdf
N2  - Temporal patterns of frequency-localized features in ASR. This work  investigates the use  of  frequency-localized temporal patterns of  the speech signal  for  developing  robust  front-end  for  Automatic  Speech  Recognition  (ASR). Various linear transforms are investigated  for  parameterization of  the frequency-localized temporal  patterns. We show  that  temporal patterns  closely  follow  the  properties  of a  first- order Markov  process,  which  results in  the PCA transforms being very close to the DCT transform.  Better  recognition performance is achieved on  using  the DCT components of temporal patterns as opposed to directly using temporal patterns  for  feature estimation. Other  linear  transforms  such  as Linear  Discriminant  Analysis  (LDA)  are  also  studied by the parameterization.  The parameterized TempoRAl  Patterns (TRAPS) are used  to estimate broad-phonetic class-posteriors independently in  each critical-band.  These class- posteriors are  combined and  used  as the features for  word  recognition.  Our work  shows that broad-phonetic features generalize better than other conventional features and yield considerable complementary information concerning short-term central features in ASR.
PB  - Oregon Health and Sciences University
PY  - 2003-07-01
T1  - Temporal patterns of frequency-localized features in ASR
TI  - Temporal patterns of frequency-localized features in ASR
UR  - https://digitalcollections.ohsu.edu/record/2416/files/3154_etd.pdf
Y1  - 2003-07-01
ER  -