The LIUM (Laboratoire d'Informatique de l'Université du Maine) have released they're work under CeCILL (GPL Compatible). They have made acoustic models, language models, and dictionnaries for sphinx 3.x, and it can be downloaded here . How could it be use with other engines ?
The Lium site has amazing French resources for Sphinx.
The site contains a French pronunciation dictionary, French Language Model (which you should be able to convert to ASCII for use with HTK - see this thread for more info). The license seems to be a BSD style license, and is thus compatible with the GPL.
The French Acoustic Models are for the CMU Sphinx 3.x decoder series (They do not work very well on Sphinx 4). Theoretically you can convert Sphinx Acoustic models to HTK/Julius acoustic models. There are some partial solutions (see this thread for more info), but there is no complete solution.
That is why we need the "source" audio used to create our own acoustic models - these acoustic models were likely created with a proprietary speech corpus, and the source audio cannot be freely distributed.
This is the reason why VoxForge was created.
Please consider submitting some speech in French (or English). Go to project Gutenberg (French), create some prompts, and record yourself reading them and submit them to us. Or send us some prompts that we can include in the Java Speech Submission applet. Any help is welcome and appreciated!