Publication Details
Effect of normalization on TRAP based systems in ASR
normalization, noisy speech recognition, TRAP, feature extraction
Effect of different kinds of normalization on noisy speech recognition based on TRAP.
Good performance in noisy conditions is one of the most important
attributes of automatic speech recognition (ASR) system in practical
applications. We are focusing on the effect of the on-line mean and
variance normalization of temporal vectors in this work. We are comparing
the performance of the Temporal Pattern (TRAP) system with different kinds of
normalization in different noisy conditions. The performance of system
with temporal vectors projected on basis obtained by linear
discriminant analysis (LDA) is also investigated. Experiments show
better performance of system using vector projection in noisy conditions.
@INPROCEEDINGS{FITPUB7284, author = "Franti\v{s}ek Gr\'{e}zl", title = "Effect of normalization on TRAP based systems in ASR", pages = "128--131", booktitle = "Proc. 13th International scientific conference Radioelektronika 2003", year = 2003, location = "Brno, CZ", publisher = "Department of Radioelectronics FEEC BUT", ISBN = "80-214-2383-8", language = "english", url = "https://www.fit.vut.cz/research/publication/7284" }