Sphinx: Open Source Speech Recognition

Have you been frustrated with speech recognition software in the past? I've used the top proprietary products, such as IBM's ViaVoice and Nuance's Dragon Naturally Speaking in the past. I've found limited good uses for them, but they're not entirely accurate, and they're reasonably expensive. Carnegie-Mellon University developed a free offering called Sphinx, which may be a good alternative for many people, and it's now offered under a BSD license.

You can either use Sphinx for straight speech recognition tasks, or integrate it with, say, an online application. Sphinx-3 is one freely downloadable version, Sphinx-4 is a versioin written entirely in Java, and PocketSphinx--though smaller and less accurate than other versions--is designed to run in real time on handhelds and be integrated with live applications.

If you want to try Sphinx for your own use, it looks like the package to try is Sphinx-4--the Java version. Because it's Java, it's platform independent, I'm still in the process of testing it for accuracy, and will do a subsequent post on my results, but this post from Artificial Intelligence and Robotics reports that the Java version works as well as proprietary solutions. It's interesting to see an open source effort in this space, where the proprietary players have always ruled.