Introduction

This page provides information on Fine-Tracker: a computational model of human word recognition, built using techniques from the field of automatic speech recognition (ASR), which is able to capture and use fine-grained acoustic-phonetic variation during speech recognition.

Like its cousin SpeM (Scharenborg et al., 2005), Fine-Tracker can take realistic representations of the speech signal as its input and subsequently perform a word search based on the theory of human word recognition as explained in Norris (1994). Fine-Tracker takes multi-tier vector representations of the speech signal, for instance created by an artificial neural network, as its input. The output of Fine-Tracker consists of an Nbest list of the most likely word sequences.

Download the Fine-Tracker (ftracker) software package

• NEW 04/11/2009: A new version of Fine-Tracker is available. Fine-Tracker Version 1.2 is able to use unigram and bigram language models. This package also includes a tool to create your own language models for use with Fine-Tracker.
• Version 1.0

More information

Details about Fine-Tracker's implementation, its objective, background, and initial experiments are described in the following papers/abstracts:

Scharenborg, O. (2009). Using durational cues in a computational model of spoken-word recognition. Proceedings of Interspeech, Brighton, UK, September 2009, pp. 1675-1678. [.pdf]

Scharenborg, O. (2008). Modelling fine-phonetic detail in a computational model of word recognition. Proceedings of Interspeech, Brisbane, Australia, September 2008, pp. 1473-1476. [.pdf]

Scharenborg, O. (2008). Fine-phonetic variation in a computational model of word recognition. Poster presentation at Acoust. Soc. Am. mtg, Paris, July 2008. [.pdf]

ftracker 1.0, 1.2 Copyright (C) 2008, 2009 Radboud University Nijmegen

Ideas by: Odette Scharenborg and Louis ten Bosch
Implementation by: Frank Kusters and Albert Gerritsen.