Throat Microphone Speech Recognition using MFCC

The Throat Microphone (TM) is a non-acoustic device, relying on the vibrations of vocal folds rather than the audible sound produced. Correctly capturing vocal fold vibrations is difficult due to poor signal representation capabilities. The system recognizes the TM vibrations and produces the corresponding speech sound. This is done by extracting features from the spectrum of the TM vibrations and comparing the obtained features with the values stored in a database. The extracted features include characteristic features of the speech waveform called Mel-Frequency Cepstral Coefficients (MFCC). The selection of the closest speech signal is chosen by the minimum mean square error estimation method, where the signal in the database whose corresponding MFCC values show the least difference from the input speech MFCCs is selected. This system has the potential of having applications for giving voice to those with defective speech and in military communications.