Carnegie Mellon University
file.pdf (118.35 kB)
Download file

Data Selection for Speech Recognition

Download (118.35 kB)
journal contribution
posted on 2008-07-01, 00:00 authored by Yi Wu, Alexander RudnickyAlexander Rudnicky, Rong Zhang

This paper presents a strategy for efficiently selecting informative data from large corpora of transcribed speech. We propose to choose data uniformly according to the distribution of some target speech unit (phoneme, word, character, etc). In our experiment, in contrast to the common belief that “there is no data like more data”, we found it possible to select a highly informative subset of data that produces recognition performance comparable to a system that makes use of a much larger amount of data. At the same time, our selection process is efficient and fast.