tfgenre - Vlady Veselinov

Download and unzip trained models in root folder (2GB)

To evaluate accuracy of all models:

sh evaluate.sh

To train all models

sh train.sh

Optional audio data extraction guide with Essentia (macOS only)

Data set used for this project can be found here

1. Install Essentia with Homebrew: brew tap MTG/essentia && brew install essentia

2. Make a data/audio folder in the project root containing subfolders of audio files. Subfolder names used as categories, i.e. data/audio/car_horn/11251.wav

3. Run src/extract_features.py with Python2.7 to load audio from data/audio and produce JSON files with audio features

4. Run write_csv.py to distill all JSON to one .csv

Optional Run TensorBoard for visualisations:

Download and unzip log files in root folder

Start TensorBoard

sh tensorboard.sh