The algorithm Vocale identifies automatically boundaries of vocalic and consonantal intervals in speech signal. To perform this task the algorithm measures the regularities of each column of the spectrogram. This is done by a hierarchical procedure which first evaluates the relative entropies of the renormalized coefficients of adjacent columns. This identifies most of the regular patterns displayed by vocalic intervals. Voiced consonants and glides are identified in a second step, using different criteria.
The code of the algorithm is written in Perl. It uses as input the spectrogram produced by Praat. Praat has been developed by Paul Boersma and David Weenink at the Institute of Phonetic Sciences of the University of Amsterdam, The Netherlands.
The algorithm Vocale is being developed as part of FAPESP's Project "Rhythmic patterns, parameter setting and language change", CNPq's project "Probabilistic tools for pattern identification applied to linguistics" and PRONEX grant 66.2177/1996-6 "Núcleo de Excelência Critical phenomena in probability and stochastic processes".
The developing team is composed by Jesus Garcia (University of São Paulo) as the main developer, Ulrike Gut (University of Bielefeld) as Phonetic expert, Flaviane Fernandes and Janaisa Viscardi (Unicamp) as Phonetic assistants. The team is coordinated by Antonio Galves.
Click here to download the algorithm.
Jesus Garcia, Antonio Galves:
Automatic identification of vocalic intervals in speech signal
Available as PowerPoint Presentation (2.731.008 bytes) and PDF (2.989.944 bytes).
Previous: Syntax and the "Tycho Brahe" Corpus
Next: Phonology and the "Sotaq" Computer Program