Implicit linguistic structure in connected speech  

Francisco Lacerda, Lisa Gustavsson & Nina Svärd

In the Proceedings of Fonetik 2003, the XVI Swedish Phonetics Conference, Lövånger, June 2-4, 2003. Phonum 9: 69-72.


This paper sketches a model of early emergence of basic linguistic structure using general-purpose similarity measures, without a priori linguistic knowledge, to structure natural speech signals. The model attempts to mimic a first language learning situation. Crude auditory representations of the acoustic signal are continuously stored in memory and processed to detect similarities between portions of the representation patterns. The similarity measures are purely auditory and allow only for moderate time warping and frequency shifts, picking up any best matches among the stored patterns and between these and incoming speech. An example of the model performance is presented.

View pdf-version of paper

PERILUS | All issues | 2003 issue | Previous | Next | Phonetics at Stockholm University