ai_industry_project

Multi-label audio file classification

Preprocessing

Steps and description can be found in feature_extraction

Models

• Music Tagging transformer (Full implementation of transformer)
• Wav2Vec transformer (finetuning model)
• Short-chunk CNN
• RNN
• (Naive predictor for baseline metrics)

Structure

• configs: configuration files to set parameters for preprocessing and the models
• data: contains preprocesed data and labels
• images: images used in markdown of the notebooks
• lib_util: different utilities that were used
• (models: saved instances of trained models, not uploaded to git)
• Plots: plotted results of preprocessing, data analysis and training
• wav_data: raw data, audio files in WAV format
• archive: old, experimental implementations, unused

Dependencies

use pip install -r requirements.txt