AVA Speech Dataset Downloader

Toolkit for downloading the audio files from the AVA Speech dataset.

AVA Speech Overview

The dataset is proviveded by Google and you can download the main csv file, our explore the dataset, here. Google also published a paper about the dataset and you can read it here.

Setup

First of all you need to download the dependecies to use the code. You can do that running the following code:

pip install -r requirements.txt

It's recommended to use virtualenv or conda virtual environment before installing the dependencies.

How to Use

To run the code is very straight forward, you just need to do:

python3 main.py

There is some optinal parameters you can use as you need:

• -labels_file or --labels_file: The main csv ("ava_speech_v1.csv") path. If none, the file will be downloaded and saved in the folder root path. Default: None.
• -fs or --fs: The frame sample (or sample rate) in Hz (8000 or 16000) which you want to save the videos. Default: 16000.
• -o or --o: The output path where you want to save the downloaded videos. If none, will be created a folder named "dataset" on the folder root path. Default: None.
• -multiprocessing or --multiprocessing: Use it if you want to use multiprocessing to speed up the process.
• -overwrite or --overwrite: Use it if you want to overwrite the files that have already beed downloaded.
• -c or --c: The class(es) you want to download ('clean-speech', 'speech-music', 'no-speech', 'speech-noise'). You can pass just one class or more. See the examples section below to see how to use. Default: all classes will be downloaded.
• -max_files or --max_files: The maximum files for each class that you want to download. Default: None, all files will be downloaded.
• -channels or --channels: How many channels the output file will have. Default: 1 (Mono channel).

Examples

Download the videos from a specific classes:

python3 main.py --c speech-noise

or

python3 main.py --c speech-noise speech-music clean-speech

Using multiprocessing to speed up the process:

python3 main.py --multiprocessing

Overwritting existing files that already have been downloaded:

python3 main.py --overwrite

Passing the label_csf folder

python3 main.py --labels_file {FOLDER_ROOT_PATH}

Download a max amount of videos from each specific classes:

python3 main.py --c speech-noise speech-music --max_files 10

Using everything together:

python3 main.py --c speech-noise speech-music --fs 8000 --max_files 100 --multiprocessing --overwrite

In this case I want to download 100 videos from the "speech-noise" and "speech-music" classes, resample than to have a frame sample of 8000 Hz, use multiprocessing to speed up the process and overwrite all the files (if any files have already been downloaded).

Project's Structure

ava-speech-downloader
├── src
|   └── core.py
├── __init__.py
├── LICENCE
├── main.py
├── setup.py
├── requirements.txt
└── README.md

Author

• Rafael Greca