Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Data-Driven Lexical Modeling of Pronunciation Variations for ASR

Qian Yang, Jean-Pierre Martens

ELIS-RUG, Gent, Belgium

In this paper a method for the automatic construction of a lexicon with multiple entries per word is described. The basic idea is to transform a reference word transcription by means of stochastic pronunciation rules that can be learned automatically. This approach already proved its potential (Cremelie & Martens, 1999), and is now brought to a much higher level of performance. Relative reductions of the word error rate (WER) of 20 % (open vocabulary) to 45 % (closed vocabulary) are now within reach.


N. Cremelie, J.P. Martens. "In search for better pronunciation models for speech recognition," Speech Communication 29, 115-136, 1999.

Full Paper

Bibliographic reference.  Yang, Qian / Martens, Jean-Pierre (2000): "Data-driven lexical modeling of pronunciation variations for ASR", In ICSLP-2000, vol.1, 417-420.