tiktok.sounds

entity · one row = one sound

Every sound seen on a video: title, artist, owner, originality, commercial rights, TikTok matching, Spotify/Apple ids, video count.

• Coverage: Target ~1.5B sounds. ~26% of licensed commercial tracks never resolve and have no video count (is_retired).
• Refresh: After each daily sound pass; at most ~1 day old.
• Sorted by sound_id. Filters on the leading columns are fastest.

Example

SELECT *
FROM tiktok.sounds
LIMIT 100

Columns

Column	Type	Description
sound_id	UInt64
title	String
artist	String
album	String
created_at	DateTime
duration_s	UInt16	SECONDS (videos are ms)
language	LowCardinality(String)	‘English’, ‘non_vocal’, …
owner_creator_id	UInt64	original sounds only
owner_username	String
owner_nickname	String
is_original	UInt8	made on TikTok by a creator
is_official	UInt8	is_pgc: a label/distributor upload
is_commercial	UInt8	cleared for business use
is_commercial_strict	UInt8
has_custom_title	UInt8	TikTok’s own “the creator gave it a name” signal for originals. The strongest artist-discovery filter measured (median 28k videos vs 1).
has_vocals	Nullable(UInt8)	NULL = TikTok did not say
has_strong_beat	UInt8
has_lyrics	UInt8
theme_tags	Array(LowCardinality(String))
matched_type	LowCardinality(String)	How TikTok matched the audio: ‘not_found’ | ‘fingerprint’ | ‘pgc’. For finding UNDISCOVERED artists, a match is a NEGATIVE signal.
matched_song_title	String
matched_song_artist	String
spotify_id	String	dsp platform 3
apple_music_id	String	dsp platform 1
origin_video_id	UInt64	the video an original was lifted from
sim_group_id	UInt64	TikTok’s “same audio” cluster; use it to avoid double counting re-uploads
loudness_lufs	Nullable(Float32)	NULL = not measured
cover_uri	String
play_url	String	unsigned mp3, does not expire
video_count	UInt32	videos using it (TikTok’s user_count)
video_count_date	Date	when that was read
priority	UInt8	1 = video_count > 3 (daily), else 2
is_retired	UInt8	never resolves in bulk (3 misses)
first_seen	DateTime
refreshed_at	DateTime