video2spectrogram
About
This package is meant to automate the process of extracting audio files from videos and saving the plots computed from these audio frequencies in the Mel scale (Sectrogram). Videos are processed in parallel with the audio extracted by ffmpeg stored in .wav files which are then used to create spectrograms stored as .JPEG and can be used by any audio-based method.
Currently supported video formats include .mp4,mpeg-4,.avi,.wmv. If you have a different extension, you can simply change the script to include them (in the video2spectrogram/get_spectrogram.py)
Package requirements
librosanumpymatplotlib
Make sure that the above packages are installed before running any functions.
ffmpeg: You will need to have installed ffmpeg in order to perform the audio extraction from the video files.
Multiprocessing: The code uses multiprocessing for improving speeds, thus the total time required for the conversion varies across different processors. The code has been tested on an AMD Ryzen 3950X with an average conversion time of 4 minutes for ~1K videos (with an average resolution of 480p and length of 5s.)
Dataset structure
The package assumes a fixed video dataset structure:
│
└──
│ │
│ │───
│ │───
│ │─── ...
│ ...
│
└───
│ │ │ │───
│ │───
│ │─── ... ... ...
Usage
The main code is at the get_spectrograme.py file. To run the convertor simply call the convert function with the base directory of the dataset and the destination directory for where to save the audio. Additional arguments that can be used:
verbose_lvl: Integer for verbosity.save_wav: Boolean to determine if the created wav files are to be stored and not deleted.ar: Integer for the ffmpeg option for specifying the audio sampling frequency.res_h: Integer for the height of the spectrogram image to be saved.res_w: Integer for the width of the spectrogram image to be saved.dpi: Integer for the display's dot's per inch. Needs to be set to avoid inconsistencies to theresargument.
from video2spectrogram import convert
#or
from get_spectrogram import convert
convert(my_dataset_dir, my_target_dir)
Installation through git
Please make sure, Git is installed in your machine:
$ sudo apt-get update
$ sudo apt-get install git
$ git clone https://github.com/alexandrosstergiou/video2spectrogram.git
$ cd dataset2database
$ pip install .
You can then use it as any other package installed through pip.
Installation through pip
The latest stable release is also available for download through pip
$ pip install video2spectrogram
8.2k Dec 26, 2022
6 Apr 6, 2022
5 Jun 26, 2022
5 Dec 31, 2021
72 Aug 1, 2022
3 Jun 22, 2022
1 Feb 2, 2022
15 Nov 10, 2022
38 Dec 23, 2022
2 Aug 03, 2022
10k Jan 08, 2023
4 Oct 09, 2022
5.7k Jan 06, 2023
6 Dec 31, 2021
10 Sep 29, 2022
2 Apr 04, 2022
7 Aug 22, 2022
10 Dec 28, 2022
19 Nov 10, 2022
119 Dec 27, 2022
3 May 04, 2022
61 Dec 29, 2022
331 Dec 27, 2022
60 Dec 17, 2022
10 Oct 03, 2022
2 Feb 25, 2022
82 Jan 03, 2023
648 Jan 01, 2023