The number of nutrition-related diseases rose significantly in the last centuries. A precise and timesaving protocol method for nutrition monitoring is needed. In contrast, today's monitoring of food consumption is done manually by information sampling on questionnaires. A promising kind of automatic nutrition monitoring is the analysis and classification of human chewing sounds. In this work, we present the approach of using a signal recognition software tool for analysis of chewing sounds in human nutrition. We developed a sensor system consisting of miniature electret microphones integrated in a hearing aid case, an analog preprocessing stage and a notebook. For our work we generated a database of accumulated 13 hours sound data with 30 participants eating samples of seven food types and consuming one drink. Every labeled chew event (jaw opening and closing) belongs to a predefined class. Divisions of chew events are made based on the change of spectral sound energy composition with the progress of food crushing. With features from spectral analysis, Hidden Markov Models (HMMs) were trained. To classify a whole intake cycle, we used a finite state grammar (FSG) decoder based on the Viterbi algorithm. The FSG models the event sequences of an intake cycle for every food type. We achieved reasonable recognition accuracy values by evaluation on our dataset.