By short time domain processing techniques, correlation characteristics of the speech signal can be obtained. Today, how to take advantage of short-term autocorrelation function of extracting a voice signal pitch.
So, what is pitch it? Each vocal cords open and close a period of time called the pitch or pitch period, which is called the inverse of the fundamental frequency, referred to as the pitch. The length of the pitch and individual vocal cords, thickness, toughness, stiffness and pronunciation habits related, in large part reflects the individual characteristics. In addition, the pitch also with the person's sex, age may be, older men is low (about 50Hz), children and young women is high (about 450Hz). The pitch is mainly used in low bit rate speech coding, speech analysis and synthesis, speech recognition and speaker recognition, occupies a very important position in the field of voice signal.
Short-term autocorrelation function formula:
short autocorrelation function has a number of features:
1) when k takes 0, the function maximum value at this time is the short-time autocorrelation function of the signal energy (see previous article) ;
2) If the original signal sequence is a period for the period T, then the autocorrelation function is a periodic function of period T. With this characteristic, the speech signal can be calculated in the pitch.
For chestnut:
The figure is based on the length of the sampling rate, 44100Hz 0.9 second speech signal frame length is set 1200, a frame shift of 600, taking the red frame (vocal part) of the one, as shown in FIG.
From the chart (b), after removing the first maximum value (0), a maximum value at k = 236, then the frame rate corresponding to the fundamental frequency:
In addition, short-time autocorrelation function may also be used for endpoint detection, a judgment is voiced or unvoiced speech and so on. Well, today's content talk so much, see the next issue!
