Song difficulty rating (auto-computed)

[LoggeL]

Problem

Users don't know how hard a song is before committing to it. Some songs have rapid-fire lyrics, wide vocal ranges, or complex rhythms that make them poor choices for casual singers.

Proposal

Auto-compute a difficulty score (1-5 stars or Easy/Medium/Hard) for each processed song, based on objective metrics from the data we already have.

Metrics (all derivable from existing lyrics.json + audio)

1. Words per second — average across the song. Fast songs are harder.
2. Lyrics density — ratio of singing time to total duration. Songs with few breaks are harder.
3. Word complexity — average word length or syllable count. Longer words are harder to read in time.
4. Gap consistency — variance in timing gaps between words. Irregular rhythms are harder to follow.
5. Speaker changes — songs with many speaker switches (duets) are harder for solo singers.
6. Transcription confidence — low confidence often means unusual pronunciation or fast delivery.

Scoring Formula

Weighted combination of the above, normalized to a 1-5 scale:

• 1-2: Slow, clear, regular rhythm (ballads, simple pop)
• 3: Average pop/rock
• 4-5: Fast rap, complex rhythms, duets, low-confidence lyrics

Storage

• Add difficulty field to lyrics.json (alongside avg_confidence)
• Compute during post-processing in helpers.py
• Backfill existing songs (similar to confidence backfill)

Display

• Star rating or color-coded badge in library cards
• Filter/sort by difficulty in library
• Show in admin song detail view
• Show subtly in the player (next to song title)

Complexity

Low. Pure computation from existing data, no new services needed. The interesting part is tuning the formula weights.