This paper proposes a demonstration system capable of recognizing command words with a custom optopalatographic device that measures speech articulation. Beyond the measuring device, the proposed system is composed of two interacting computer programs - one for recording and one for the actual word recognition. Operating offline, the recording software can be used to record large datasets, which in turn are used by the recognition software to train and validate classification models. Operating online, the recording software can be used to record single words, which are classified using pre-trained classification models. The system was developed modularly, enabling different devices for measuring speech articulation and different classification models to be easily integrated. The current classification system employs dynamic time warping in combination with a pattern matching algorithm and achieves 98.38% accuracy on a single speaker 40-word corpus.