On the back explains how to pitch with a short analysis of the autocorrelation function of the speech signal in. Recalling the core formula under short-term autocorrelation function:
As can be seen from the above equation, short-term autocorrelation function relates the multiplication, the total number of frames with each frame fn is increased, the more clearly the calculated short-time autocorrelation function details, the better pitch detection, the computing volume increases. Today, we introduce another pitch detection algorithm, which uses the difference multiplication algorithm substituted, thereby greatly reducing the amount of computation.
AMDF short average magnitude difference function (Average Magnitude Difference Function)
After the speech signal x (n) sub-frames, each data frame short average magnitude difference function is shown below:
From the above equation, if the period of the speech signal is M, then when k = 0, ± M, when ± 2M ......, Di (k) = 0. Thus, short-time average magnitude difference function pitch to be detected.
For chestnut:
The figure is a speech signal frame, separately subjected to short-time autocorrelation function analysis and short-time average magnitude difference function analysis, the results as shown below:
Visible, short-term autocorrelation function and the short-time average magnitude difference function are closely related, short-term autocorrelation function peak corresponds to the short-time average magnitude difference valley function.
In principle, in accordance with the short-time average magnitude difference function formula, the magnitude of the integer multiple of the sample period should be the same, i.e. Di (k) = 0, but the result was the above-described non-zero. The reason is that the voice signal is not a mere periodic signal, only occurs in a valley at the pitch period by short-time average magnitude difference function, and increase over time, it will diminish the depth of the valley.
Written in the last, time-domain speech signal processing related article is mainly known by Song compiled the "Application of MATLAB in speech signal analysis and synthesis," a book, according to my knowledge the first book written in MATLAB simulation software in verify, and record the results for everyone to share, but also to facilitate their access at any time during the review.
