wavio

| Audio Fingerprinting Without the ML Tax
Most audio identification tools today reach for embeddings and neural nets. wavio doesn't.

It's a fast, deterministic acoustic fingerprinting library written in Rust — built on the same peak-based approach as Shazam, with none of the ML overhead. No embeddings, no models, no runtime. Just spectral peaks, combinatorial hashing, and raw speed.

How it works

wavio runs a straightforward DSP pipeline:

Each step is deterministic: the same input always produces the same fingerprint. That makes results reproducible and debuggable — no model drift, no version mismatches.

Why it's fast

• In-memory & on-disk indexing — query thousands of tracks in microseconds
• Zero unsafe code (#![forbid(unsafe_code)])
• Optional parallelism via rayon

Benchmarks on synthetic 22,050 Hz audio (release build):

Task	Median
Fingerprint a 3-min track	88.6 ms
Index 1,000 tracks	1.04 ms
Query (1,000 lookups)	0.57 µs/query

Use it from Rust, Python, or the CLI

wavio ships as a Rust crate, a Python package (via PyO3/maturin), and a CLI:

wavio-cli index --db ./wavio.db ./music/
wavio-cli query --db ./wavio.db ./clip.wav

Who it's for

If you're building music identification, duplicate detection, or content matching and don't want to carry an ML stack for it, wavio gives you a lean, predictable alternative.

Check it out: github.com/MinLee0210/wavio