STFT
===
librosa: https://librosa.org/doc/0.8.0/generated/librosa.stft.html

Basic:
---
- **sample rate**: How many points in a second.
- if sr=1000, the data can represent at most 500 Hz's sound.

STFT Steps
---
1. Take n points as a **Frame**
- ex. Each Frame has 512 points

    |  Frame 1  |
    |-----------|
     ^512 points^

2. The next Frame will has an offset from first one.
- ex.

    | Frame 1 | ---------
    ---| Frame 2 | ------
    ------| Frame 3 | ---

3. Do FFT at each Frame

4. Result:

    Frequency
    |     |     |
    |  F  |  F  |
    |  r  |  r  |
    |  a  |  a  |
    |  m  |  m  |
    |  e  |  e  |
    |     |     |
    |  1  |  2  |
    |     |     |
    |_____|_____|____[Magnitude][M][M][M] -> represent time

Figures:
---
- Each Frame:

Magnitude
|
|
| /\
| / \
| / \
|_/______\_______Frequency

- each_Frame.Transpose combine:

Frequency
| | |
| F | F |
| r | r |
| a | a |
| m | m |
| e | e |
| | |
| 1 | 2 |
| | |
|_____|_____|____[Magnitude][M][M][M] -> represent time

    
    
google colab
---
reference: 
https://github.com/weichian0920/MFA_DAE/blob/main/src/utils/signalprocess.py
https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9167389

google colab:
https://colab.research.google.com/drive/1wsap4JQarLkTajfuQ2g4v7T-oA4_t-9x?usp=sharing