Thank you to all of the contributors who have keept CMUSphinx alive over the last decade, in particular to Nickolay Shmyrev and the whole team at Alpha Cephei. As you may have noticed, active development has mostly ceased over the last few years, and the technological foundation of CMUSphinx has become quite antiquated.
For state-of-the-art speech recognition the Alpha Cephei team is now working exclusively on Vosk, and there are a number of other open source options, notably Coqui, wav2vec, Julius, TensorFlowASR, DeepSpeech and of course Kaldi.
Nonetheless, there are still many people using CMUSphinx and PocketSphinx in particular, so there is some value in maintaining (if not actually developing) it. Its users frequently encounter difficulties due to the build system, which could be corrected by modernizing the codebase slightly. Due to the eternal "pre-alpha" status of the system, there are also many problems of portability and stability that should be adressed.
For this reason, we are preparing a true release of PocketSphinx, with a focus on a modern build system with no external dependencies, and a stable, documented, and easy to use API in C and Python. In addition, SphinxTrain will either continue to be maintained or will simply be integrated into PocketSphinx.
Finally, SourceForge is no longer a viable option for hosting. From now on, the GitHub Project is the official home of CMUSphinx, and we will soon migrate all the other downloads (models, etc) and close the SourceForge site.