On-device music identification for a video production pipeline.
Vidsy MusicMatch UI
Context
At Vidsy, creators send finished videos with music, voiceover, and sound effects baked into a single audio track. Before editing can begin, someone has to manually listen through and identify which supplied music track was used.
It's slow, repetitive, and one of several manual steps I've been working to automate.
The bigger picture is a full production pipeline that handles audio separation, footage matching, text extraction, and edit reconstruction.
I'm breaking that down into standalone components, building and proving each one individually.
MusicMatch was the first: can you instantly identify a baked-in music track without relying on cloud APIs or racking up costs?
Music Match macOS app icon
Role
- Problem identification
- Product design
- macOS development (Swift, SwiftUI, ShazamKit)
- UI Design
Stress testing the tool with 7999 tracks
Solution
- Built a native macOS app that takes two inputs: a creator's video or audio file and a folder of supplied music tracks.
- Accepts local files or a direct platform link, downloading the video automatically if needed.
- Uses ShazamKit to fingerprint the audio and match it against every track in the library, including nested subfolders.
- Returns a ranked list of results with confidence percentages and a Reveal in Finder button for the matched track.
- Runs entirely on-device. No internet, no API keys, no cost per use. Local machine learning, instant results.
- UI is deliberately minimal: two drop zones, a link input, a progress bar, results. Nothing else.
Music Match uses ShazamKit under the hood
Impact
A task that used to involve manual listening and guesswork now takes seconds.
The motion design team's response was immediate and enthusiastic. It's a small, focused utility that does exactly one job, and it does it well. It's also the first shipped proof that the larger automation pipeline works in practice, one component at a time.