Inside the HumMatch Matching Engine: Introducing the HumMatch API

Every music app can tell you what is popular. Almost none of them can tell you what you, specifically, can sing.

That second question is the one HumMatch was built to answer, and answering it well turns out to require something most of the music industry has never bothered to collect: measured vocal data. Today we want to show you, at a high level, how our matching engine thinks, and announce something we have been working toward for a while: the HumMatch API, so other products can build on it too.

The data underneath

The engine starts with two datasets we have been building since day one.

Songs. The HumMatch catalog now covers more than 14,000 songs, each with a researched vocal range: the actual span of notes the lead vocal asks for, sourced and confidence-rated rather than guessed from genre. That is the difference between "this is a pop song" and "this chorus lives between G3 and G4 and never asks for more."

Voices. Alongside the catalog, we maintain a growing database of measured artist vocal profiles, built from analysis of real recordings. These go well past a simple low-note/high-note pair: they capture tessitura (the band where a voice actually spends its time), timbre characteristics (the measurable texture that makes two voices with the same range sound completely different), and voice embeddings: numerical fingerprints that let us compare voices the way a map compares locations.

When you hum three notes into HumMatch, you are creating your own entry in that second dataset: a Vocal ID that can be compared against every song and artist profile we have.

What the match score actually weighs

A naive matcher checks one thing: is the song's highest note below your highest note? That test passes far too many songs that will hurt and fails the question that matters: will this song feel good to sing?

Our match score is multi-factor. Without opening the hood all the way, the big ideas are:

• Comfort-zone overlap. How much of the song sits inside the band where your voice lives (not the extremes you can technically reach, but the range you can sit in all night).
• Penalties for unsingable notes. Notes above your ceiling or below your floor do not just lower a score politely. They are the difference between a song you perform and a song you survive, and the engine treats them that way.
• Center distance. How far the song's melodic center of gravity sits from the center of your comfortable range. A song can overlap your comfort zone and still spend most of its time at the uncomfortable end of it.
• Similarity in voice-space. Using those measured artist profiles, the engine can ask a subtler question: how close is your voice, as a point in a high-dimensional space, to the voice that made this song famous? Two voices that cluster together tend to flatter the same material.

A worked example: two songs, both "in range"

Say a singer's full range runs E3 up to C5, and their comfort zone (where the voice actually lives) is A3 to A4. Now consider two songs:

• Song A sits at E4–C5. Technically every note is inside the singer's range.
• Song B sits at G3–G4. Also entirely inside the singer's range.

A range-only matcher shrugs: both pass. Our engine strongly prefers Song B, and the numbers explain why.

On the musical number line (where each step is one semitone), the comfort zone A3–A4 spans 12 steps. Song B's span, G3–G4, overlaps that comfort zone for 10 of its 12 semitones, about 83% of the song lives where the voice lives. Song A overlaps for only 5 of its 8 semitones, and the part that does not overlap is the worst part: its top notes run right up to C5, the singer's absolute ceiling. That means every chorus is sung at the very edge of what is physically possible. No headroom, no margin for a long night or a second take.

That is the difference between "in range" and "in your voice." Song A is a dare. Song B is a home game.

Introducing the HumMatch API

Here is the announcement: the matching engine is no longer only ours. The HumMatch Partner API exposes the same vocal-fit logic that powers hummatch.me, as clean, stateless endpoints:

• Vocal profiles. Send a vocal range (note names like "C3"/"A4", or raw MIDI numbers) or hum-derived data, and get back a normalized vocal profile: voice type, tone label, and a reusable Vocal ID.
• Song fit. Score any song in our catalog against a profile and get a real match score, plus the same plain-English "why it fits" explanation HumMatch shows singers on the site.
• Group fit. Send two or more profiles and get back the group's shared vocal range with ranked candidate songs, the logic behind SquadMatch, computed entirely on the data you send.

Every call is a real-time computation plus a read-only catalog lookup. We do not store what you send beyond the API key itself.

Free for builders, enterprise for platforms

The free tier is live today. Developers and hobbyists can self-serve a sandbox key from the developer page in one request. No signup form, no approval queue, key issued instantly, with sensible rate limits. If you have ever wanted to add "can this user actually sing this?" to a project, you can have it working this afternoon.

The enterprise tier is for products. Karaoke platforms that want smarter song queues. Music apps that want recommendations grounded in a listener's actual voice. Vocal coaching tools that want measurable range and progress data. Catalog owners who want to know which audiences can sing their songs. For live keys, higher limits, and integration support, we are actively opening partnership conversations. Reach out through our contact form and mention API access.

Where this goes next

The roadmap from here is about depth. As the platform grows, we plan to pursue deeper catalog integrations and licensed data expansions: richer song-level vocal data, broader coverage across languages and genres, and closer relationships with the people who own and manage catalogs. The direction is simple: HumMatch wants to be the vocal-fit layer for anywhere music meets a human voice.

It started with humming three notes in a browser. It is becoming infrastructure.