How our engineers taught an EQ to read a room

Every track has a fingerprint. Genre, tempo, vocal density — Vibe reads all of it in under 200 ms and loads the right curve before the first beat lands. Here's how we built that.

The problem with manual EQ

Most people never touch their EQ. Not because they don't care about sound — they do — but because the gap between 'I want more bass' and 'boost 80 Hz by 3 dB' is too wide. Manual EQ is a tool for engineers. We wanted something that just worked.

The idea was simple: what if the app knew what you were about to hear and set itself up before the first beat? Not after 30 seconds of playback. Before the drop.

What a fingerprint actually is

When a track starts, Vibe grabs its metadata — title, artist, album — and cross-references it against a multi-source lookup. Genre tags, tempo, harmonic content, vocal presence. This happens locally, on your phone, in around 180 ms.

The result isn't a number. It's a profile. A jazz trio at 100 BPM with sparse low-end needs a completely different curve than a hyperpop track at 178 BPM with a distorted 808. The model knows the difference.

The presets aren't random

We didn't pick 8 presets because it's a round number. Each one was tuned for a specific listening context — not a genre, a vibe. Afterhours is for rooms with sub-bass you can feel. Focus is flat because when you're working you want accuracy, not color. Hype is a V-shape because the gym is a V-shape.

The 6 hidden presets are weirder. They unlock based on what you actually listen to. Use Vibe for three weeks and you'll notice new options appearing. We won't spoil them.

Why it works offline

The fingerprinting stack runs entirely on-device. No API call, no cloud dependency. This was a deliberate call — we didn't want Auto-EQ to break because you're underground or because a third-party service is down.

The tradeoff is app size, which is why STELLARvibe is 118 MB. That's the voice engine and the classification model living on your phone. Worth it.

"The goal was zero conscious effort. You press play, it sounds right. That's it."

— Luca Ferri, lead audio engineer

What's next

We're working on real-time adaptation — not just at track start, but as the song evolves. A track that opens sparse and builds to a wall of sound should have an EQ that follows it. That's a harder problem. We're on it.