audiologic 0.1.1

Python utilities to process and predict on audio attributes

Python Module to process and predict on music attributes

Two Models were built and trained to predict valence given an audio sample. One uses a feature pipeline on top of librosa to make a number of predictors that go into a Random Forest model to determine a valence prediction. The other uses OpenAI's whisper model to transcribe lyrics, then tokenize the words, and again a trained Random Forest model makes the prediction based on lyrics.

	Model RMSE
Audio	1.56
Lyrics	1.28

Data Used:

• 1000 Song Dataset - Download here
• Spotify Developer API - 30 second previews

Package Requirements

pip install -r requirements.txt

• make sure to download whisper from openai (not currently included in requirements.txt)
• Also must install ffmpeg (using brew, choco, etc.)