Project 04 / 15

TrackPulse

AI-powered Audio Intelligence to classify vocal vs. instrumental, detect language and flag explicit content.

Case study / 04

Behind the build

The problem, approach, and impact of TrackPulse.

Type: AI Audio Analysis
For: Music catalog, distribution and licensing teams that screen batches of audio tracks.

Reviewing audio one track at a time slows cataloging and content checks, while running local speech models can require extra infrastructure.

Built a web interface and API flow that processes WAV files from an uploaded ZIP and streams per-track analysis results to the browser.

Classifies whether a track contains vocals, identifies language and flags explicit content.
Streams progress and individual results as newline-delimited JSON, so completed tracks appear without waiting for the entire batch.
Validates and normalizes model output before display and retries selected transient API failures.

Lets users review a batch as results arrive.
Processes ~5min/track, 20 tracks in 5 minutes with ⬆90% classification accuracy.

Engineering / Technologies

Technologies and tools powering this project.