Audio Tagging

Upload an audio file and get AudioSet labels (527 classes) from a PANNs Cnn14 model.

Drop an audio file here or click to choose
wav, mp3, m4a, flac, ogg, webm …

Top K

How many labels to show, best first (527 possible)

Min probability

Hide labels scoring below this (0 = show all)

Per-segment (sec)

Also tag each N-second window (0 = whole file only)

Script (optional, .pdf or .txt)

With a script the table follows it: real character names, scripted lines placed in time, stage directions snapped to detected sounds.

Speakers (optional)

How many people speak. Used when no script is given; fixes over-splitting.

Timestamped timeline (when things happen, second by second)

Breakdown table (speech turns + events, with interpretation)

Production-style table: one row per speech turn and per notable sound event, in time order, with transcript, speaker, entity and an interpretation column.

Transcript table (speech turns with what was said)

One row per stretch of speech: when it ran, the transcribed words, and what else was audible (music, other sounds). Adds noticeably to processing time.

Uses a frame-level model (~10ms resolution) to say when a sound starts and stops, e.g. music from 12s–90s or an impact at 45.3s. It only names generic sound classes (Music, Speech, Slam, Thud, ...), not what caused them.