Struct BeatThis 

pub struct BeatThis<M: Model> { /* private fields */ }

High-level beat tracker composing the full pipeline.

Owns the mel spectrogram model, the beat prediction model, and the peak picker. Generic over the model type, so it works with any backend (ort, rten, tract).

Implementations

impl<M: Model> BeatThis<M>

pub fn new<R: Runtime<Model = M>>( runtime: &R, mel_model_path: &Path, beat_model_path: &Path, ) -> Result<Self>

Create a new beat tracker by loading both ONNX models via the given runtime.

• runtime: any Runtime (e.g. OrtRuntime::default())
• mel_model_path: path to the mel spectrogram ONNX model
• beat_model_path: path to the beat tracking ONNX model

pub fn from_models(mel_model: M, beat_model: M) -> Self

Create a beat tracker from pre-built models.

Use this when you need separate runtimes for the mel and beat models (e.g. one with profiling enabled).

pub fn beat_model_mut(&mut self) -> &mut M

Get a mutable reference to the beat prediction model.

Useful for runtime-specific operations like ending ORT profiling.

pub fn analyze_audio( &mut self, samples: &[f32], sample_rate: u32, ) -> Result<BeatAnalysis>

Run the full pipeline on raw audio samples.

The samples are resampled to 22050 Hz if sample_rate differs. Input should be mono f32 PCM.

pub fn analyze_audio_timed( &mut self, samples: &[f32], sample_rate: u32, ) -> Result<TimedAnalysis>

Run the full pipeline on raw audio samples, returning per-stage timing.

pub fn analyze_file(&mut self, path: &Path) -> Result<BeatAnalysis>

Run the full pipeline on an audio file.

Loads the file, resamples to 22050 Hz mono, computes mel spectrogram, runs beat prediction, and decodes into beat/downbeat timestamps.