July / Monthly / Tutorials / Software / Home


The Institute for Signal and Information Processing (ISIP) was created in 1994 to develop public domain speech recognition software. One of the primary goals of our program is to educate students and researchers who are new to speech recognition, in addition to providing tools for those already established in the field. Therefore, ISIP offers speech recognition software with freely available state-of-the-art tools to support the development of public research and technology.

The development of educational tutorials is not a new undertaking for ISIP. One of our earlier tutorials, constructed by Aravind Ganapathiraju, guided users through a step-by-step production of a speech recognizer that could decode continuous alphadigit strings. Now ISIP has embarked, once again, on releasing an upgraded, more powerful version of our recognition software. The latest release of the software is accompanied by complete descriptions of all phases of the speech recognition process in a single Tutorial Book. This on-line book provides a theoretical overview for each phase as well as detailed explanations of how to accomplish that phase using our software. We have been meticulous in our efforts to provide descriptions useful to both novice and advanced users.

The Tutorial Book contains a section for each phase of the speech recognition process, as outlined below:

 
 
Installation
Explains how to install the ISIP software.

Data Preparation
Explains how to prepare data for feature extraction.

Feature Extraction
Describes how to numerically measure the features of a person's speech and store the measurements in feature vectors.

Acoustic Modeling
Explains how to train word models and phone models that are extracted from the numerical measurements.

Language Modeling
Explains how to generate network and N-gram models.

Recognition
Describes how to perform the recognition process.

Tutorial Examples
Describes, through selected examples, how to take you from the beginning to the end of the speech recognition process using ISIP's software.

 
 

The first three sections of the Tutorial Book are available for review. The latter sections are still under construction.

Organization

The Tutorial Book contains a main section for each topic listed above. The Table of Contents allows easy access to any of these sections. This provides novices a starting point and a structure to follow in order to learn about each phase of the speech recognition process. It also allows advanced users direct access to specific topics of interest. The Table of Contents page is shown below.



Each section listed in the Table of Contents is linked to an Overview page of that section. For example, the image below is an example of the overview page from 3. Feature Extraction.



The Overview page contains a brief description of the section topic along with a second-level Table of Contents from that section. Each heading and subheading of the Table of Contents is hyperlinked to the appropriate pages. Again, the organization allows advanced users to go directly to topics of interest while providing novice users an outline to follow.

Navigation

Each page of a section contains a navigation bar, located in the upper right hand corner.

The navigation bar enhances usability of the Tutorial Book, providing an additional method for accessing information. Five buttons located in the bar permits the following modes of navigation through each section.

Takes you directly to the Tutorial Book's Table of Contents.

Takes you directly to the Overview page of the section you are currently working.

Allows you to move forward a page.

Allows you to back a page.

Moves up one heading level at a time.

To support our goal of providing easy access to information for both novice and advanced users we give step-be-step instructions for many processes. Examples of such instructions include how to install the ISIP software and how to prepare data for feature extraction. Also, we explain all steps in training acoustic models and preparing language models for specific applications such as TIDigits and Switchboard.

We hope you enjoy using our software. If you have any questions concerning the Tutorial Book or if any problems arise with your experiments, feel free to e-mail us at ies_help@cavs.msstate.edu. You will receive an immediate reply.