Audio analysis (fallback)
Audio details
| Essentia | Deezer | MIRFLEX | |
|---|---|---|---|
| Mood | — | — | — |
| Tempo | — | — | — |
| Key | — | — | — |
| Vocals | — | — | — |
| Singer | — | — | — |
| Danceable | — | — | — |
| Instrument | — | — | — |
| Auto-tags | — | — | — |
| Loudness | — | — | — |
| Duration | — | — | — |
| Explicit | — | — | — |
Essentia mood / tempo / key / vocals / singer-gender / danceable are extracted with Essentia's
pre-trained models (MTG-Jamendo, MSD-MusiCNN, MTT, Discogs Effnet) running on the track's 30-second Deezer preview. Deezer tempo / loudness / duration / explicit-flag come
from the Deezer /track API — when BPMs
disagree by a factor of two (e.g. 87 vs 173), it's tempo-detection ambiguity, not a different recording.