Expand description
Streaming BeatNet beat/downbeat tracking without Python or an async runtime.
Construct and process on a worker thread. Only transport::AudioProducer
belongs in an audio callback. Frame/event times refer to the source audio,
not wall-clock delivery time; the frontend needs 706 samples of lookahead.
Re-exports§
pub use model::Inference;pub use model::Model;pub use model::OnnxModel;pub use tracker::TrackerConfig;
Modules§
- features
- Exact BeatNet/madmom frontend, with centered frames and bounded lookahead.
- model
- Synchronous ONNX inference with explicitly retained recurrent state.
- tracker
- Bounded Rust port of BeatNet’s particle-filter cascade.
- transport
- Bounded callback-to-worker transport. Payloads are
Copyand have no destructors.
Structs§
- Beat
Event - An accepted beat on the source audio timeline.
- BeatNet
- Synchronous PCM processor.
processaccepts arbitrary interleaved chunk sizes, including chunks ending partway through a multichannel frame. - Beat
NetConfig - Source audio format, pretrained model, and tracker settings.
- Frame
Output - State and optional beat event produced together for one feature frame.
- Tracking
State - Posterior estimates for a single 20 ms feature frame.