Life Sound Signal Acquisition, Feature Parameter Extraction And Modeling Based On Bone-conducted Microphone

Eating habits, no matter healthy or not, are directly related to people’s daily lives.With the continuous development of medical standards and scientific andtechnological level, people are increasingly concerned about health problems. Mobilemedical care, emerging as the time requires, monitors the oral motor status. This kindof behaviour not only effectively supervises modern citizens’ eating habits, but alsodetects a problem in time, which is crucial to the improvement of people’s healthawareness.In this paper, we do some work in the life sound signals acquisition, whichmainly includes three parts: the first is determining the position of the bone-conductedmicrophone device, the second is creating the bone-conducted sounds corpus, and thelast one is marked the signals and analysis the data.In the aspect of extracting signals characteristic parameter, according to thecontribution rate of each band, we arrange the distributing of filters, and then wepropose a new method of life sound adaptive frequency scale transformation, weverify the correctness of our method in the signal recognition.In the field of signal modeling, we built a dynamic model on peoples’ diet signals(especially chewing signals) by using the features of n-gram language model. Whichleads to the improvement of chewing signal recognition rate that ensuring the highaccuracy of voice signal identification.