Speech Analysis and Processing for Knowledge Discovery Aalborg, Denmark June 4-6, 2008

Incorporating Suprasegmental Knowledge For Phone Recognition With Conditional Random Fields

Prateeti Mohapatra, Eric Fosler-Lussier

Department of Computer Science and Engineering, The Ohio State University, Columbus, OH, USA

In this paper, we investigate the integration of lexical stress and syllabic position of segments in a Multi-Layer Perceptron (MLP) classification system that is part of our Conditional Random Fields (CRF) phone recognizer. CRFs are used to integrate MLP posterior estimates, particularly of phonological features or phonetic classes, which stand in as representations of the acoustics; we show that incorporating suprasegmental information as part of the MLP classification system augments the acoustic space in a beneficial way for phonological feature based CRF models. TIMIT phone recognition experiments show a small but statistically significant improvement.

Full Paper
Presentation

Bibliographic reference.  Mohapatra, Prateeti / Fosler-Lussier, Eric (2008): "Incorporating suprasegmental knowledge for phone recognition with conditional random fields", In SPKD-2008, paper 054.