Digital voice Signal Processing learning notes--short time-frequency domain analysis of speech signals (2)

4.3 Interpretation of the filter

1. Short-time Fourier transform filter Realization form One

watermark/2/text/ahr0cdovl2jsb2cuy3nkbi5uzxqvam9qb3poyw5nanu=/font/5a6l5l2t/fontsize/400/fill/i0jbqkfcma==/ Dissolve/70/gravity/southeast ">

watermark/2/text/ahr0cdovl2jsb2cuy3nkbi5uzxqvam9qb3poyw5nanu=/font/5a6l5l2t/fontsize/400/fill/i0jbqkfcma==/ Dissolve/70/gravity/southeast ">

2. Short-time Fourier transform filter Realization form two

Make:

The

4.4 Time domain and frequency domain sample rate in short time spectrum

the short-time Fourier transform is a two-dimensional representation of one-dimensional signal. That

watermark/2/text/ahr0cdovl2jsb2cuy3nkbi5uzxqvam9qb3poyw5nanu=/font/5a6l5l2t/fontsize/400/fill/i0jbqkfcma==/ Dissolve/70/gravity/southeast "> is a function of time n and angular frequency W at the same time. How to restore the origin

watermark/2/text/ahr0cdovl2jsb2cuy3nkbi5uzxqvam9qb3poyw5nanu=/font/5a6l5l2t/fontsize/400/fill/i0jbqkfcma==/ Dissolve/70/gravity/southeast "style=" font-size:18px ", the first encounter is the time domain sample rate and frequency domain sample rate problem.

1. Time Domain Sample rate

The sample rate required within the time domain is. B is the bandwidth of the window, the length of the N form sequence, and FS is the sample rate of the signal x (n).

2. frequency domain Sample Rate

In order to recover from it.

watermark/2/text/ahr0cdovl2jsb2cuy3nkbi5uzxqvam9qb3poyw5nanu=/font/5a6l5l2t/fontsize/400/fill/i0jbqkfcma==/ Dissolve/70/gravity/southeast "style=" font-size:18px "> must use the following set of frequency values to sample

3. Total sample Rate

We are able to determine what is necessary to make the original signal non-aliasing in each second.

watermark/2/text/ahr0cdovl2jsb2cuy3nkbi5uzxqvam9qb3poyw5nanu=/font/5a6l5l2t/fontsize/400/fill/i0jbqkfcma==/ Dissolve/70/gravity/southeast "style=" font-size:18px ">" The total number of samples. The minimum sample rate in the time domain is 2B, and B is the band bandwidth of the window. and the smallest sample in the frequency domain is n. is the window width. So

watermark/2/text/ahr0cdovl2jsb2cuy3nkbi5uzxqvam9qb3poyw5nanu=/font/5a6l5l2t/fontsize/400/fill/i0jbqkfcma==/ Dissolve/70/gravity/southeast "style=" font-size:18px "> Total sample Rate (SR) equals

watermark/2/text/ahr0cdovl2jsb2cuy3nkbi5uzxqvam9qb3poyw5nanu=/font/5a6l5l2t/fontsize/400/fill/i0jbqkfcma==/ Dissolve/70/gravity/southeast ">

In most real windows, B can be represented as a multiple of fs/n. That

watermark/2/text/ahr0cdovl2jsb2cuy3nkbi5uzxqvam9qb3poyw5nanu=/font/5a6l5l2t/fontsize/400/fill/i0jbqkfcma==/ Dissolve/70/gravity/southeast ">

c is a proportional constant, which is:

In the formula. Sr/fs is the "over-rate sample ratio" that is obtained compared to the general sampling frequency. Hamming window 2c=4, rectangular window 2c=2.

Digital voice Signal Processing learning notes--short time-frequency domain analysis of speech signals (2)