Recognition of speech from signals secondary to speech

In two separate studies, a correlation has been observed between the myoelectric signals (MESs) present in neck muscles during speech and the acoustic waveform. In the first study, five experiments were conducted using a 17-word vocabulary. The application of maximum likelihood pattern recognition to the averaged myoelectric signal produced recognition accuracies as high as 98% for a two-word set with accuracy settling at five times a priori. The second study used a five-word subset of the original vocabulary. Graphical results were obtained to display the action of the MES during the acoustic signal. The results of both studies indicate that the MES contains sufficient information to develop a pattern recognition scheme to decode the spoken word (phoneme) from its corresponding MES.<>