Introduction

Welcome! This guide will take you through all the basics of building a speech recognizer with the ICSI speech recognition tools. By the end of working through these pages, you will have seen how to go from a set of training sound files to a working recognizer, and you will be familiar with all the programs and data formats involved. This will take you a couple of hours, although it's easy to split it up into several sessions.

Assumptions

- software, filesystem etc.

Contents

The guide continues with the following pages:

• The overview gives a picture of the entire process and introduces the important terms and concepts.
• Standard training takes you through the basic sequence involved in building a new recognizer for a speech database.
• Embedded training explains a modified training procedure, in which the training labels and the classification network are iteratively refined.
• Real-time operation
• Other issues - corpora
• Glossary

About this guide This guide was started in May 1997 as a reworking of an original Guide to ICSI Speech Recognition produced by Dan Jurafsky in 1996. The managing author is Dan Ellis, with the invaluable assistance of ... and the rest of the ICSI realization group.