Struct WaveNetModel 

pub struct WaveNetModel<const CH: usize, const K: usize, const HEAD: usize> {
    pub array1: WaveNetLayerArray<1, 1, CH, K, HEAD>,
    pub array2: WaveNetLayerArray<CH, 1, HEAD, K, 1>,
    pub head_scale: f32,
    pub receptive_field_size: usize,
    pub prewarm_on_reset: bool,
}

Complete WaveNet Model containing Two Heterogeneous Layer Array Blocks.

Scientific Reference: van den Oord, A., et al. (2016). “WaveNet: A Generative Model for Raw Audio.” DeepMind.

CH = Array1 channels (layer 0 of the JSON, e.g., 16 for Standard) K = kernel size (always 3) HEAD = Array1 head_size = Array2 channels (e.g., 8 for Standard)

Array2 uses HEAD channels and projects to 1 output (HEAD2=1), following the C++ pattern: WaveNetLayerArrayT<CH, 1, 1, HEAD, K, Dilations, true>.

Fields

array1: WaveNetLayerArray<1, 1, CH, K, HEAD>

Inner array 01: IN=1, COND=1, CH channels, HEAD outputs, no HeadBias.

array2: WaveNetLayerArray<CH, 1, HEAD, K, 1>

Inner array 02: IN=CH, COND=1, HEAD channels, 1 output, with HeadBias.

head_scale: f32

Final voltage compensation scale (Target Output Scale).

receptive_field_size: usize

Largest circular buffer required at the Kernel’s temporal root.

prewarm_on_reset: bool

Whether to execute prewarm during reset(). Default: true.

Implementations

impl<const CH: usize, const K: usize, const HEAD: usize> WaveNetModel<CH, K, HEAD>

pub fn set_effective_layers(&mut self, n: usize)

Sets the effective number of layers on both arrays for soft-degrade.

pub fn backup_buffer_starts(&self, starts: &mut [usize], offset: &mut usize)

Backs up the buffer_start pointers of both layer arrays into a slice.

pub fn restore_buffer_starts(&mut self, starts: &[usize], offset: &mut usize)

Restores the buffer_start pointers of both layer arrays from a slice.

pub fn process(&mut self, input: &[f32], output: &mut [f32])

Resolves the full forward pass and produces waveform samples in zero allocation (DSP).

Combines the outputs of both arrays: sum(head1) + sum(head2) × head_scale.

For Scientists and Devs: The dispatch_simd! macro monomorphizes this function via the SimdMath trait, matching the global hardware configuration once at the call site. The compiler generates branchless assembly optimized for your processor (AVX2, AVX-512, or AVX-512 VNNI BF16) without v-table or dynamic dispatch overhead.

pub fn prewarm(&mut self)

Stabilizes the model by processing silence (Zero Input) for pre-warm.

Dispatches via dispatch_simd! macro — a single static match on the globally detected SIMD_MATH.instruction_set — to run the AVX2, AVX-512, or AVX-512 VNNI BF16 kernel without runtime feature detection per call.

pub unsafe fn prewarm_avx512(&mut self)

Prewarm strictly optimized for AVX-512 architecture.

Safety

Requires a supported processor (AVX-512).

pub unsafe fn prewarm_avx2(&mut self)

Prewarm strictly optimized for AVX2 architecture.

Safety

Requires an x86-64-v3 (AVX2) processor.

Trait Implementations

impl<const CH: usize, const K: usize, const HEAD: usize> NamModel for WaveNetModel<CH, K, HEAD>

fn process(&mut self, input: &[f32], output: &mut [f32])

Invoked by the DSP RT-Thread to process acoustic sample blocks (Float32).

fn prewarm(&mut self, _num_samples: usize)

“Heats up” the virtual tubes of the neural engine (prewarm).

fn prewarm_samples(&self) -> usize

Returns the number of samples needed to fully stabilize the model’s internal state (receptive field / recurrent memory depth).

fn prewarm_on_reset(&self) -> bool

Returns whether prewarm should be executed on reset().

fn set_prewarm_on_reset(&mut self, val: bool)

Sets whether prewarm should be executed on reset().

fn reset(&mut self, _sample_rate: u32, max_buffer_size: usize) -> Result<()>

Resets the model’s internal state with a new sample rate and max buffer size.

fn set_max_buffer_size(&mut self, _max_buf: usize) -> Result<()>

Reallocates internal buffers to support the given maximum block size.

fn slimmable_breakpoints(&self) -> Vec<f64>

Returns the breakpoints at which slimmable quality transitions occur.