Crate pitch_core 

Streaming pitch (f0) tracker. Pure-DSP backends only — neural backends live in the companion crate pitch-core-onnx.

Quick start

use pitch_core::{PitchTracker, SwipeEstimator};

let est = SwipeEstimator::new()?;
let mut tracker = PitchTracker::new(est, 48_000, 1024)?;

for frame in tracker.process(&chunk)? {
    if frame.confidence > 0.3 {
        println!("{:.3}s  {:.1} Hz", frame.time_s, frame.pitch_hz);
    }
}

Adding ONNX backends

Add pitch-core-onnx as a dependency and pass any of its estimators to the same PitchTracker::new. The trait surface is identical:

use pitch_core::PitchTracker;
use pitch_core_onnx::{SwiftF0Estimator, Mode};

let est = SwiftF0Estimator::new("path/to/swift_f0.onnx", Mode::Balanced)?;
let mut tracker = PitchTracker::new(est, 48_000, 1024)?;

Re-exports

pub use estimator::calibrate_confidence;

pub use estimator::EstimatorError;

pub use estimator::PitchEstimator;

pub use estimator::PitchFrame;

pub use estimator::Result;

pub use praat_ac::PraatAcEstimator;

pub use pyin_est::PyinEstimator;

pub use swipe::SwipeEstimator;

Modules

estimator

praat_ac

Praat-style autocorrelation pitch estimator (Boersma 1993).

pyin_est

pYIN (probabilistic YIN, Mauch & Dixon 2014) via the pyin crate (Sytronik, pure-Rust port of sevagh/pitch-detection).

resample

swipe

SWIPE’/SWIPE pitch estimator — thin adapter over the public swipe-rs crate (Apache-2.0). The pitch detection algorithm itself is implemented in https://crates.io/crates/swipe-rs; this file just plugs it into pitch-core’s PitchEstimator trait and adds the is_preliminary: false field that swipe-rs (deliberately) doesn’t know about.

Structs

PitchTracker

High-level streaming pitch tracker. Combines any PitchEstimator with a linear resampler that converts the host’s sample rate to the estimator’s target rate.