<aside> ❄️ Autotag is a tool designed for the automatic classification of music collections using deep learning techniques (Essentia / TensorFlow).

</aside>

Use

• Audio files are loaded and processed sequentially.
• For each file, several attributes are extracted and classified.
• Results are then exported in csv/json format for analysis or further use.

Detail

• Autotag distinguishes between simple attributes (instrumental or vocal, sad or happy, dancing or not, tempo, ...) and multiple-choice attributes (musical genres, instrument types, emotions).
• For multiple-choice attributes, predictions are averaged and the best values are selected as Top5 or Top10.

Processing

• Audio Analysis : We use the Essentia library for the initial processing of audio files. It loads and transforms audio files into formats compatible with deep learning models.
• Deep Learning : TensorFlow is an open-source machine learning framework running pre-trained deep learning models. These models recognize and classify various musical features from audio files.
• Specific models (Discogs-EffNet, MSD-MusiCNN, TempoCNN) are used to analyze characteristics such as genre, tempo, mood, theme, and other relevant attributes.