Since the late 1980's, we have promoted a vision of free software for speech and signal processing research following in the footsteps of other successful projects, such as the Free Software Foundation. Our goal has been to develop a comprehensive software environment for performing data intensive science and engineering research that is flexible yet computationally efficient. Our premier application is large vocabulary speech recognition though our software is designed to be much more general than this specific application.

Our approach to creating such a lasting infrastructure incorporates a diversity of technologies, including some novel Internet-based Java applets that let users interact with the technology without the need to install the software locally. We also host an extensive signal processing web site that includes on-line courses, tutorials, and toolkits. Support, training, and education are important ingredients to our success.

The goal in this presentation is two-fold. First, we review the challenges associated with developing such a large-scale application. Second, we discuss the limitations of current programming paradigms. Though software tools and concepts have advanced significantly in the past decade, there are still some very fundamental obstacles that preclude the development of simple, easy-to-use, and extensible software tools for engineers.