For segmentation of the words' data from the sound data that we are receiving from the microphone, we are using a simple algorithm. We find the average of the amplitude of the input data over a sampling window (i.e. over a small time interval) compare it with a threshold value. As soon as the average becomes greater than the threshold, we assume the word to have started. And as soon as the average falls below another threshold level, which is lower than the earlier threshold, we assume the word to have ended. In this way we segment the word data into an array which is different from the input array as we do not include the quiet periods. The threshold values were chosen by hit and trial, taking into account the ambient noise levels.
Graphs of words
Back