Unlocking Audio Network's Catalog with Lyrics

Production music lives or dies by its discoverability. A music editor cutting a documentary or an agency building a creative campaign rarely arrive knowing exactly which track they want. Instead, they are guided by a feeling, a scene, or a line of copy. The faster a catalog can help solve that puzzle, the more likely its music will end up in the final cut.

For instrumental music, search has long been driven by genre, mood, instrumentation, and tempo. But for vocal tracks, the single richest search signal is the one most catalogs are missing: the lyrics.

The challenge: a vocal catalog without lyric data

Across a catalog of this scale, many older vocal recordings were never delivered to Audio Network along with the track. For Audio Network, that created three gaps:

Search. Audio Network's creators, music editors, and clients couldn't search the catalog by what a song actually says. A brief calling for "songs about starting over" or a spot that needed a specific turn of phrase had no reliable way to surface matching tracks.

Delivery. Lyrics are also a deliverable. Clients licensing production music increasingly expect lyric data along with the track and its stems, for clearance review, captioning and accessibility, localization, or their own downstream search systems.

Easier editing & accessibility. Timing matters as much as the words themselves. A text lyric file can tell you a track mentions "the long way home"; but word-level alignment — meaning, the karaoke-style, word-by-word timestamping of the track — tells you that that phrase happens at 1:47. Alignment is what makes lyrics usable for caption timing, lyric-synced playback, clip-level search, and matching a specific line to a specific moment on screen.

With thousands of tracks missing lyrics across the Audio Network catalog, manual transcription wasn’t realistic. Doing it by ear, track by track, would have taken years.

The solution: automated lyric transcription at catalog scale

Audio Network used AudioShake's lyric transcription and word alignment API to generate time-stamped lyrics for its catalog.

AudioShake's approach is built on source separation. Rather than transcribing a finished mix — where vocals compete with drums, guitars, and orchestration — AudioShake uses its state-of-the-art vocal model to first isolate the vocal from the recording, then transcribes the isolated stem, providing the transcription model a “clean” input to process.

The service takes an audio file as input and returns transcriptions in seconds, in multiple formats, including time-aligned output that allows a precise pinpoint of where a word corresponds to each moment in the track. AudioShake’s industry-leading approach to lyrics transcription has led it to lead the benchmarks leaderboards — significantly outperforming OpenAI’s Whisper, Music AI, OWSM, and other industry models.

The result

Lyric search across the catalog. Audio Network's creators and music editors can now search on lyrical content alongside genre, mood, and instrumentation, matching briefs to tracks on meaning as well as sound.

Lyrics delivered with the music. Lyric data now ships with the tracks Audio Network licenses, giving clients what they need for review, captioning, localization, and their own search and metadata workflows.

Deeper catalog reach. Tracks that were effectively invisible to a lyric-driven brief are now discoverable — which puts more of the catalog in front of more clients.

Key capabilities used by Audio Network

• Lyric transcription
• Vocal separation
• Time-aligned lyric output (“word alignment”)
• Batch catalog processing via API