Struct StreamingBackendSession 

pub struct StreamingBackendSession { /* private fields */ }

A reusable stateful denoising session for continuous planar audio.

Supported backends retain their overlap, recurrent model, and resampler state between calls. Model-backed sessions load and optimize their graph at construction and never reopen it while processing or resetting the stream.

Implementations

impl StreamingBackendSession

pub fn supports(backend: Backend) -> bool

Return whether a compiled backend has a continuous stateful adapter.

pub fn new( backend: Backend, sample_rate: u32, channels: usize, denoiser: DenoiserConfig, backend_options: BackendOptions, ) -> Result<Self, String>

Construct a stream and allocate every backend state before accepting audio. backend_options must already contain any managed model path.

pub fn new_with_accelerator( backend: Backend, sample_rate: u32, channels: usize, denoiser: DenoiserConfig, backend_options: BackendOptions, accelerator: AcceleratorSelection, ) -> Result<Self, String>

Construct a stream using an already-resolved accelerator snapshot.

This keeps capability reporting, recipe identity, and model preparation bound to the same decision when a frontend preflights the runtime before opening an input or audio device.

pub fn new_gtcrn_with_prepared_model( sample_rate: u32, channels: usize, denoiser: DenoiserConfig, backend_options: BackendOptions, model: &GtcrnModel, ) -> Result<Self, String>

Create a GTCRN stream from an already authenticated and optimized graph.

The options still undergo the ordinary deterministic/runtime validation; callers must prepare model for the resulting effective runtime.

pub fn new_gtcrn_for_daw( sample_rate: u32, channels: usize, denoiser: DenoiserConfig, backend_options: BackendOptions, ) -> Result<Self, String>

Construct the GTCRN stream at the wider, bounded DAW host-rate limit.

This is format-neutral and retains the ordinary backend, resource, and accelerator checks. Only file/offline sample-rate validation is replaced by the plug-in host contract.

pub fn new_gtcrn_for_daw_with_prepared_model( sample_rate: u32, channels: usize, denoiser: DenoiserConfig, backend_options: BackendOptions, model: &GtcrnModel, ) -> Result<Self, String>

Create a prepared GTCRN stream at the wider, bounded DAW host-rate limit.

This is format-neutral and retains the ordinary backend, model, resource, and accelerator checks. Only file/offline sample-rate validation is replaced by the plug-in host contract.

pub fn backend(&self) -> Backend

pub const fn accelerator(&self) -> AcceleratorSelection

pub fn estimate_additional_bytes( backend: Backend, sample_rate: u32, channels: usize, channel_mode: ChannelMode, ) -> Result<u64, ConfigError>

Conservative backend-specific state beyond the classical stream and caller-owned input/output blocks.

pub fn estimate_vad_additional_bytes( sample_rate: u32, channels: usize, block_frames: usize, frame_size: usize, profile_ms: f64, ) -> Result<u64, ConfigError>

Conservative VAD alignment state beyond the backend and ordinary caller-owned stream blocks.

pub fn process_block( &mut self, channels: &[Vec<f64>], ) -> Result<Vec<Vec<f64>>, String>

Process a block. The returned block can be empty while bounded model or sample-rate-converter latency is retained.

pub fn finish(&mut self) -> Result<Vec<Vec<f64>>, String>

Flush every pending model frame and converter delay exactly once.

pub fn reset(&mut self) -> Result<(), String>

Start an independent stream while retaining any already-loaded model.