This is an attempt to train a CMU Sphinx acoustic model for Portuguese (pt-br) using the datasets found in an ASR study for Brazillian Portuguese. The resulting model is used in Rhasspy.
Following the sphinxtrain tutorial, a continuous 200 senome acoustic model was train on approximately 8 hours of speech data. No attempt was made to validate or clean the data, since I don't know Portuguese. The trained model does poorly, as expected on so little data, but may be useful enough for a very constrained domain.