Applied AI projectAudio · NLP · Scoring
AI Music Scoring
An integrated Python pipeline that analyzes audio and lyrics to produce a genre-aware music quality score from 0 to 100.
Project record
This page reports the maturity and evidence documented in Ethosoft's working archive. It does not imply peer review, regulatory clearance, or clinical validation.
Methods and evidence
AI Music Scoring
An integrated Python pipeline that analyzes audio and lyrics to produce a genre-aware music quality score from 0 to 100.
- Vocal and lyric pipeline
- Demucs separates vocals, Whisper transcribes them, and deterministic rhyme/redif features are extracted from the resulting lyrics.
- Language analysis
- DistilBERT estimates semantic depth while logistic regression provides a documented content-toxicity signal.
- Audio analysis
- MFCC and spectral features feed Gradient Boosting models for acousticness, energy, and danceability estimates.
- Scoring
- Genre-specific coefficients combine the numeric features into a 0–100 score; the project document reports MAE and F1 as evaluation metrics. Read the project PDF.
Evidence boundary: the supplied project document describes the methodology and reports strong model performance, but does not provide a public benchmark, annotator protocol, or independent validation of the final score's alignment with listener judgment.