Articulatory-feature-based confidence measures

Leung, Kim Hung; Siu, Man-Hung

doi:10.1016/j.csl.2005.08.003

Cited by 3 publications

(2 citation statements)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Alternately, an ASR engine can produce many types of scores which are used as observations to train a statistical model. In addition to a typical ASR observation such as acoustic score, decoders may produce a variety of observations based on the language model, articulatory observations [3] or discourse events [4]. The framework of observing events from an output (lattice or otherwise) to train a model for estimating confidence is also used in the fields of information extraction [5] and machine translation [6,7].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Maximum Entropy Confidence Estimation for Speech Recognition

White¹,

Droppo

Acero

et al. 2007

2007 IEEE International Conference on Acoustics, Speech and Signal Processing - ICASSP '07

View full text Add to dashboard Cite

For many automatic speech recognition (ASR) applications, it is useful to predict the likelihood that the recognized string contains an error. This paper explores two modifications of a classic design. First, it replaces the standard maximum likelihood classifier with a maximum entropy classifier. The maximum entropy framework carries the dual advantages discriminative training and reasonable generalization. Second, it includes a number of alternative features. Our ASR system is heavily pruned, and often produces recognition lattices with only a single path. These alternate features are meant to serve as a surrogate for the typical features that can be computed from a rich lattice. We show that the maximum entropy classifier easily outperforms the standard baseline system, and the alternative features provide consistent gains for all of our test sets.

show abstract

Section: Introductionmentioning

confidence: 99%

“…While the recent trend has been toward discriminative systems[ 1 1, 12], many systems still train a generative model based on observations pulled from a lattice [8,3].…”

Section: Introductionmentioning

confidence: 99%