I am trying out Jasper on Raspbian. According to Jasper docs, along with PocketSphinx, we have to also install CMUCLMTK, MIT Language Modeling Toolkit, m2m-aligner, OpenFST, and Phonetisaurus.
Once I finished installing PocketSphinx, but before installing all those dependencies, I used pocketsphinx_continuous to test voice recognition. Regardless of accuracy, it obviously was able to capture my speech and went through the interpretation process. I even created my own .lm and .dict files using CMU's lmtool and tried them on pocketsphinx_continuous without problem.
My question is this: PocketSphinx seems to be able to perform speech recognition without all those dependencies. Why does Jasper tell me to install them? Does PocketSphinx really need them? What purposes do they serve?
In this list at least 2 are not needed. CMUCLMTK and MITLM are both language modeling toolkits with the same functionality. Even if you want to replace online lmtool, you need just one of them. Same, Phonetisaurus requires MITLM and m2m-aligner only to train g2p models, not to use them.
So ideally you only want MITLM, OpenFST and phonetisaurus to replace online lmtool and add new words locally. You do not need CMUCLMTK and m2m-aligner.
In the future Pocketsphinx will be able to work with new words without Phonetisaurus, so you'll be able to drop this dependency.
You need to complain to Jasper developers if they are still available for complaints.