Janus 3 Tutorial - Welcome Page

Introduction

This tutorial is intended for different levels of experience that you might have with Janus 3. If you know little about Janus 3, you might have a look at the introduction-to-Janus 3 page. It does not address the usage of Janus but talks about the basic principles used in Janus and its history and design.

For a quick reference use the tutorial index page that provides an alphabetically sorted list of items with links to their corresponding detailed description.
To jump right in to running Janus, you can choose one of the threads described at the bottom of this page.

Structure

The following diagram (it's not a click-sensitive map) shows the structure of this tutorial. There are three different threads, that address the problem of building a recognizer, namely the "discussion thread", the "scripts thread", and the "do-it-yourself thread".

The discussion thread contains detailed discussions of many things that can be done with Janus. It is useful for understanding what is going on and what the principles and ideas behind some algorithms are. It explains common problems and their solutions. There is one page which lists all these things and contains a brief description of the topic and a link to the page with the details.

The scripts thread contains all the Tcl-scripts that are used in the do-it-yourself thread. There is one page which lists all the available scripts. Each entry in this list has a brief description of what the script does and a link to the page containing the script. Since this is a tutorial, the scripts are kept as simple as possible, all unnecessary stuff like unneeded options or security checks are left out.

The do-it-yourself thread contains detailed instructions that allow you to exactly reproduce the entire development process as described in the default development scheme. Once you've understood how this works, you should be able to design your own development schemes.

If you would like to reproduce the steps in this tutorial for yourself, you can get yourself a copy of "data.tar.gz" (approx. 10MB). This archive contains an entire database (a tiny subset of the Wall Street Journal database). It contains only some raw recordings, their transcriptions, a dictionary, some simple ASCII labels, and a simple language model. Nothing else is needed to create a full blown recognizer.

Try the Typical Development Scheme for an overview of how a recognizer can be build. There is one default scheme explained in detail, which you can live through, if you like, by doing everything yourself. Each of the development steps has links to the corresponding pages of all three threads

More Information
The tutorial is not intended for explaining to source-code hackers how to implement new features or entire modules. It is also not intended to describe all functions and their argument lists; most of this information can be obtained using the interactive Janus help functions.

For questions, problems or suggestions for this tutorial contact Ivica Rogina. If you find bugs in the code or have some good ideas how to make it better write to Janus.