Windowed Frequency Power

The above figure displays the sentence, “Do you go to Ohio State?” and the corresponding power frequency plot. This plot is created by taking the DFT of the audio in a windowed section and multiplying it by its complex conjugate. These values are then plotted. Here, the word Ohio has several peaks in its power due to it being a multiple syllable word. Herein lies the problem for splitting up words by syllables. Between the peaks, if the power goes below the threshold a split will occur. This false positive doesn’t happen in silence thresholding because there isn’t a non-negligible silence between syllables in most speech.