7th International Conference on Spoken Language Processing September 16-20, 2002 Denver, Colorado, USA

Qualcomm-ICSI-OGI Features for ASR

Andre Adami (1), Lukás Burget (1), Stephane Dupont (2), Hari Garudadri (3), Frantisek Grezl (1), Hynek Hermansky (1), Pratibha Jain (1), Sachin Kajarekar (1), Nelson Morgan (4), Sunil Sivadas (1)

(1)Oregon Health & Science University, USA; (2) Multitel, Belgium; (3) Qualcomm Inc., USA; (4) International Computer Science Institute, USA

Our feature extraction module for the Aurora task is based on a combination of a conventional noise suppression technique (Wiener filtering) with our temporal processing techniques (linear discriminant RASTA filtering and nonlinear TempoRAl Pattern (TRAP) classifier). We observe better than 58% relative error improvement on the prescribed Aurora Digit Task, a performance level that is somewhat better than the new ETSI Advanced Feature standard. Furthermore, to test generalization of our approach to an independent test set not available during development, we evaluate performance on American English SpeechDatCar digits and show 10.54% relative improvement over the new ETSI standard.

Full Paper

Bibliographic reference.  Adami, Andre / Burget, Lukás / Dupont, Stephane / Garudadri, Hari / Grezl, Frantisek / Hermansky, Hynek / Jain, Pratibha / Kajarekar, Sachin / Morgan, Nelson / Sivadas, Sunil (2002): "Qualcomm-ICSI-OGI features for ASR", In ICSLP-2002, 21-24.