After a long history in leading the field in speech recognition, CMU decided to also lead in another direction by releasing its Sphinx recognizer as free software. Since the release of Sphinx II, a real-time multi-lingual, multi-platform speech recognizer, we have continued to add to this with new acoustic models for narrow band (telephone) and wideband (desktop) speech. With the release of SphinxTrain, the scripts and programs to build new acoustic models, and the release of the Cambridge/CMU Language Model toolkit, import parts of the free software speech chain are easily accessible. Sphinx III is under development offering more accurate recognition through use of acoustic models with fully continuous observation densities.