Generic TSV

Generic TSV is CVAT’s custom format for audio datasets. It only contains annotations without audio recordings.

Generic TSV export

Applicable to all audio tasks.

• Supported annotations: Intervals.
• Attributes: Supported.

Exporting an audio task or job in Generic TSV 1.0 produces a tab-separated .tsv file. The file contains these columns:

Column	Description
id	Interval identifier.
filename	Source audio filename.
subset	Source task subset name, or default if not specified.
start	Interval start timestamp.
stop	Interval end timestamp.
label	Interval label.
source	Annotation source.
score	Annotation confidence score.

CVAT adds one column for each label attribute used by the task. Audio media cannot be included in the exported dataset.

Generic TSV import

To upload interval annotations, select Generic TSV 1.0 in Upload annotations or Import annotations, then provide a tab-separated .tsv file. The file must include the filename, start, stop, and label columns:

filename	    start	        stop	        label	    transcript
recording.mp3	00:00:00.000	00:00:02.450	speech	    Hello, world.
recording.mp3	00:00:03.100	00:00:04.000	music

start and stop accept HH:MM:SS.fraction timestamps. The optional score column sets the interval confidence score. To import label attributes, add columns with names that match the configured label attributes.