Goals: Classifies niche and closely related subgenres of electronic music. Converts audio to Mel spectograms and performs image classification to determine subgenre
(Contact me to see the code, it isn't public right now: howe.gaged@gmail.com)
Progress:
- Connect Google Colab to my data
- Function to convert all MP3 files to an n-seconds-long WAV segment chosen by average loudness (LUFS) (reduced processing time and storage requirements of previous implementation by 3000%)
- Function to make a mel spectrogram from each segment with FFT magic
To Do:
- Organize data according to genre with Pandas dataframes
- Implement and train semi-supervised clustering model on existing data
- Evaluate accuracy
- Fine-tune and optimize functions and model (spectrograms are very noisy)
Tools:
- pandas
- torchaudio - transforms
- torchvision
- pydub - AudioSegment
- os
- matplotlib
