Struct BackendSession 

pub struct BackendSession { /* private fields */ }

A prepared denoising backend that can process multiple independent files.

Model-backed variants parse their source once. Fixed-shape adapters retain one optimized graph, while dynamic adapters retain the most recently required tensor shape. The session can be shared between batch workers; recurrent or per-file DSP state remains local to each process call. Loading also detaches later inference from replacement of the model pathname.

Implementations

impl BackendSession

pub fn prepare( backend: Backend, options: BackendOptions, ) -> Result<Self, String>

Validate the resolved options and load the selected backend once.

pub fn prepare_with_accelerator( backend: Backend, options: BackendOptions, accelerator: AcceleratorSelection, ) -> Result<Self, String>

Load a backend using an already-resolved accelerator selection.

Application services use this entry point so recipe identity, result reporting, and model preparation consume the same capability snapshot.

pub const fn backend(&self) -> Backend

Return the backend whose graph or state factory this session owns.

pub fn options(&self) -> &BackendOptions

Return the resolved options captured when the session was prepared.

pub const fn accelerator(&self) -> AcceleratorSelection

Return the concrete runtime captured during preparation.

pub fn process( &self, channels: &[Vec<f64>], sample_rate: u32, classical_config: &DenoiserConfig, ) -> Result<Vec<Vec<f64>>, String>

Process finite planar audio while preserving channel count and length.

This method is safe to call concurrently on a shared session. Dynamic ONNX adapters retain only their most recently requested compiled shape; an in-flight graph remains valid through reference counting.

Trait Implementations

impl Debug for BackendSession

fn fmt(&self, formatter: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.