opus1 0.2.1

//! LACE, the speech enhancer of SILK at decoder complexity 6: its layers and
//! the constants of its C header.
//!
//! Generated by `tools/dnn-gen` from libopus 1.6.1 `dnn/lace_data.h` and
//! `dnn/lace_data.c` (`init_lacelayers`). Do not edit: change the generator and
//! run it again (`tools/dnn-gen/README.md`).

// The float constants keep every digit of the C header.
#![allow(clippy::excessive_precision)]

use crate::Error;
use super::{Layer, LayerSpec, LinearSpec};
use crate::dnn::nnet::LinearLayer;
use crate::dnn::weights::WeightList;

/// The file of this model's weights in `dnn-weights/`.
pub const BLOB_FILE: &str = "lace.bin";

// The constants of `dnn/lace_data.h`, in the same order.

/// `LACE_PREEMPH`, `0.85f` in C.
pub const LACE_PREEMPH: f32 = 0.85;

/// `LACE_FRAME_SIZE`.
pub const LACE_FRAME_SIZE: usize = 80;

/// `LACE_OVERLAP_SIZE`.
pub const LACE_OVERLAP_SIZE: usize = 40;

/// `LACE_NUM_FEATURES`.
pub const LACE_NUM_FEATURES: usize = 93;

/// `LACE_PITCH_MAX`.
pub const LACE_PITCH_MAX: usize = 300;

/// `LACE_PITCH_EMBEDDING_DIM`.
pub const LACE_PITCH_EMBEDDING_DIM: usize = 64;

/// `LACE_NUMBITS_RANGE_LOW`.
pub const LACE_NUMBITS_RANGE_LOW: usize = 50;

/// `LACE_NUMBITS_RANGE_HIGH`.
pub const LACE_NUMBITS_RANGE_HIGH: usize = 650;

/// `LACE_NUMBITS_EMBEDDING_DIM`.
pub const LACE_NUMBITS_EMBEDDING_DIM: usize = 8;

/// `LACE_COND_DIM`.
pub const LACE_COND_DIM: usize = 128;

/// `LACE_HIDDEN_FEATURE_DIM`.
pub const LACE_HIDDEN_FEATURE_DIM: usize = 96;

/// `LACE_NUMBITS_SCALE_0`, `1.0983514785766602f` in C.
pub const LACE_NUMBITS_SCALE_0: f32 = 1.0983514785766602;

/// `LACE_NUMBITS_SCALE_1`, `2.0509142875671387f` in C.
pub const LACE_NUMBITS_SCALE_1: f32 = 2.0509142875671387;

/// `LACE_NUMBITS_SCALE_2`, `3.5729939937591553f` in C.
pub const LACE_NUMBITS_SCALE_2: f32 = 3.5729939937591553;

/// `LACE_NUMBITS_SCALE_3`, `4.478035926818848f` in C.
pub const LACE_NUMBITS_SCALE_3: f32 = 4.478035926818848;

/// `LACE_NUMBITS_SCALE_4`, `5.926519393920898f` in C.
pub const LACE_NUMBITS_SCALE_4: f32 = 5.926519393920898;

/// `LACE_NUMBITS_SCALE_5`, `7.152282238006592f` in C.
pub const LACE_NUMBITS_SCALE_5: f32 = 7.152282238006592;

/// `LACE_NUMBITS_SCALE_6`, `8.277412414550781f` in C.
pub const LACE_NUMBITS_SCALE_6: f32 = 8.277412414550781;

/// `LACE_NUMBITS_SCALE_7`, `8.926830291748047f` in C.
pub const LACE_NUMBITS_SCALE_7: f32 = 8.926830291748047;

/// `LACE_PITCH_EMBEDDING_OUT_SIZE`.
pub const LACE_PITCH_EMBEDDING_OUT_SIZE: usize = 64;

/// `LACE_FNET_CONV1_OUT_SIZE`.
pub const LACE_FNET_CONV1_OUT_SIZE: usize = 96;

/// `LACE_FNET_CONV1_IN_SIZE`.
pub const LACE_FNET_CONV1_IN_SIZE: usize = 173;

/// `LACE_FNET_CONV1_STATE_SIZE`, `(173 * (0))` in C.
pub const LACE_FNET_CONV1_STATE_SIZE: usize = 0;

/// `LACE_FNET_CONV1_DELAY`.
pub const LACE_FNET_CONV1_DELAY: usize = 0;

/// `LACE_FNET_CONV2_OUT_SIZE`.
pub const LACE_FNET_CONV2_OUT_SIZE: usize = 128;

/// `LACE_FNET_CONV2_IN_SIZE`.
pub const LACE_FNET_CONV2_IN_SIZE: usize = 384;

/// `LACE_FNET_CONV2_STATE_SIZE`, `(384 * (1))` in C.
pub const LACE_FNET_CONV2_STATE_SIZE: usize = 384;

/// `LACE_FNET_CONV2_DELAY`.
pub const LACE_FNET_CONV2_DELAY: usize = 0;

/// `LACE_FNET_TCONV_KERNEL_SIZE`.
pub const LACE_FNET_TCONV_KERNEL_SIZE: usize = 4;

/// `LACE_FNET_TCONV_STRIDE`.
pub const LACE_FNET_TCONV_STRIDE: usize = 4;

/// `LACE_FNET_TCONV_IN_CHANNELS`.
pub const LACE_FNET_TCONV_IN_CHANNELS: usize = 128;

/// `LACE_FNET_TCONV_OUT_CHANNELS`.
pub const LACE_FNET_TCONV_OUT_CHANNELS: usize = 128;

/// `LACE_FNET_GRU_OUT_SIZE`.
pub const LACE_FNET_GRU_OUT_SIZE: usize = 128;

/// `LACE_FNET_GRU_STATE_SIZE`.
pub const LACE_FNET_GRU_STATE_SIZE: usize = 128;

/// `LACE_CF1_FILTER_GAIN_A`, `0.690776f` in C.
pub const LACE_CF1_FILTER_GAIN_A: f32 = 0.690776;

/// `LACE_CF1_FILTER_GAIN_B`, `0.000000f` in C.
pub const LACE_CF1_FILTER_GAIN_B: f32 = 0.000000;

/// `LACE_CF1_LOG_GAIN_LIMIT`, `1.151293f` in C.
pub const LACE_CF1_LOG_GAIN_LIMIT: f32 = 1.151293;

/// `LACE_CF1_KERNEL_SIZE`.
pub const LACE_CF1_KERNEL_SIZE: usize = 16;

/// `LACE_CF1_LEFT_PADDING`.
pub const LACE_CF1_LEFT_PADDING: usize = 8;

/// `LACE_CF1_FRAME_SIZE`.
pub const LACE_CF1_FRAME_SIZE: usize = 80;

/// `LACE_CF1_OVERLAP_SIZE`.
pub const LACE_CF1_OVERLAP_SIZE: usize = 40;

/// `LACE_CF1_IN_CHANNELS`.
pub const LACE_CF1_IN_CHANNELS: usize = 1;

/// `LACE_CF1_OUT_CHANNELS`.
pub const LACE_CF1_OUT_CHANNELS: usize = 1;

/// `LACE_CF1_NORM_P`.
pub const LACE_CF1_NORM_P: usize = 2;

/// `LACE_CF1_FEATURE_DIM`.
pub const LACE_CF1_FEATURE_DIM: usize = 128;

/// `LACE_CF1_MAX_LAG`.
pub const LACE_CF1_MAX_LAG: usize = 301;

/// `LACE_CF1_KERNEL_OUT_SIZE`.
pub const LACE_CF1_KERNEL_OUT_SIZE: usize = 16;

/// `LACE_CF1_GAIN_OUT_SIZE`.
pub const LACE_CF1_GAIN_OUT_SIZE: usize = 1;

/// `LACE_CF1_GLOBAL_GAIN_OUT_SIZE`.
pub const LACE_CF1_GLOBAL_GAIN_OUT_SIZE: usize = 1;

/// `LACE_CF2_FILTER_GAIN_A`, `0.690776f` in C.
pub const LACE_CF2_FILTER_GAIN_A: f32 = 0.690776;

/// `LACE_CF2_FILTER_GAIN_B`, `0.000000f` in C.
pub const LACE_CF2_FILTER_GAIN_B: f32 = 0.000000;

/// `LACE_CF2_LOG_GAIN_LIMIT`, `1.151293f` in C.
pub const LACE_CF2_LOG_GAIN_LIMIT: f32 = 1.151293;

/// `LACE_CF2_KERNEL_SIZE`.
pub const LACE_CF2_KERNEL_SIZE: usize = 16;

/// `LACE_CF2_LEFT_PADDING`.
pub const LACE_CF2_LEFT_PADDING: usize = 8;

/// `LACE_CF2_FRAME_SIZE`.
pub const LACE_CF2_FRAME_SIZE: usize = 80;

/// `LACE_CF2_OVERLAP_SIZE`.
pub const LACE_CF2_OVERLAP_SIZE: usize = 40;

/// `LACE_CF2_IN_CHANNELS`.
pub const LACE_CF2_IN_CHANNELS: usize = 1;

/// `LACE_CF2_OUT_CHANNELS`.
pub const LACE_CF2_OUT_CHANNELS: usize = 1;

/// `LACE_CF2_NORM_P`.
pub const LACE_CF2_NORM_P: usize = 2;

/// `LACE_CF2_FEATURE_DIM`.
pub const LACE_CF2_FEATURE_DIM: usize = 128;

/// `LACE_CF2_MAX_LAG`.
pub const LACE_CF2_MAX_LAG: usize = 301;

/// `LACE_CF2_KERNEL_OUT_SIZE`.
pub const LACE_CF2_KERNEL_OUT_SIZE: usize = 16;

/// `LACE_CF2_GAIN_OUT_SIZE`.
pub const LACE_CF2_GAIN_OUT_SIZE: usize = 1;

/// `LACE_CF2_GLOBAL_GAIN_OUT_SIZE`.
pub const LACE_CF2_GLOBAL_GAIN_OUT_SIZE: usize = 1;

/// `LACE_AF1_FILTER_GAIN_A`, `1.381551f` in C.
pub const LACE_AF1_FILTER_GAIN_A: f32 = 1.381551;

/// `LACE_AF1_FILTER_GAIN_B`, `0.000000f` in C.
pub const LACE_AF1_FILTER_GAIN_B: f32 = 0.000000;

/// `LACE_AF1_SHAPE_GAIN`, `1.000000f` in C.
pub const LACE_AF1_SHAPE_GAIN: f32 = 1.000000;

/// `LACE_AF1_KERNEL_SIZE`.
pub const LACE_AF1_KERNEL_SIZE: usize = 16;

/// `LACE_AF1_FRAME_SIZE`.
pub const LACE_AF1_FRAME_SIZE: usize = 80;

/// `LACE_AF1_LEFT_PADDING`.
pub const LACE_AF1_LEFT_PADDING: usize = 15;

/// `LACE_AF1_OVERLAP_SIZE`.
pub const LACE_AF1_OVERLAP_SIZE: usize = 40;

/// `LACE_AF1_IN_CHANNELS`.
pub const LACE_AF1_IN_CHANNELS: usize = 1;

/// `LACE_AF1_OUT_CHANNELS`.
pub const LACE_AF1_OUT_CHANNELS: usize = 1;

/// `LACE_AF1_NORM_P`.
pub const LACE_AF1_NORM_P: usize = 2;

/// `LACE_AF1_FEATURE_DIM`.
pub const LACE_AF1_FEATURE_DIM: usize = 128;

/// `LACE_AF1_KERNEL_OUT_SIZE`.
pub const LACE_AF1_KERNEL_OUT_SIZE: usize = 16;

/// `LACE_AF1_GAIN_OUT_SIZE`.
pub const LACE_AF1_GAIN_OUT_SIZE: usize = 1;

/// The layers in the order of `init_lacelayers`: the arrays that each one
/// reads, and its shape. The names are the array names of the blob.
pub const LAYERS: [LayerSpec; 14] = [
    LayerSpec::Linear(LinearSpec {
        name: "lace_pitch_embedding",
        bias: Some("lace_pitch_embedding_bias"),
        subias: None,
        weights: None,
        float_weights: Some("lace_pitch_embedding_weights_float"),
        weights_idx: None,
        diag: None,
        scale: None,
        nb_inputs: 301,
        nb_outputs: 64,
    }),
    LayerSpec::Linear(LinearSpec {
        name: "lace_fnet_conv1",
        bias: Some("lace_fnet_conv1_bias"),
        subias: None,
        weights: None,
        float_weights: Some("lace_fnet_conv1_weights_float"),
        weights_idx: None,
        diag: None,
        scale: None,
        nb_inputs: 173,
        nb_outputs: 96,
    }),
    LayerSpec::Linear(LinearSpec {
        name: "lace_fnet_conv2",
        bias: Some("lace_fnet_conv2_bias"),
        subias: Some("lace_fnet_conv2_subias"),
        weights: Some("lace_fnet_conv2_weights_int8"),
        float_weights: Some("lace_fnet_conv2_weights_float"),
        weights_idx: None,
        diag: None,
        scale: Some("lace_fnet_conv2_scale"),
        nb_inputs: 768,
        nb_outputs: 128,
    }),
    LayerSpec::Linear(LinearSpec {
        name: "lace_fnet_tconv",
        bias: Some("lace_fnet_tconv_bias"),
        subias: Some("lace_fnet_tconv_subias"),
        weights: Some("lace_fnet_tconv_weights_int8"),
        float_weights: Some("lace_fnet_tconv_weights_float"),
        weights_idx: None,
        diag: None,
        scale: Some("lace_fnet_tconv_scale"),
        nb_inputs: 128,
        nb_outputs: 512,
    }),
    LayerSpec::Linear(LinearSpec {
        name: "lace_fnet_gru_input",
        bias: Some("lace_fnet_gru_input_bias"),
        subias: Some("lace_fnet_gru_input_subias"),
        weights: Some("lace_fnet_gru_input_weights_int8"),
        float_weights: Some("lace_fnet_gru_input_weights_float"),
        weights_idx: None,
        diag: None,
        scale: Some("lace_fnet_gru_input_scale"),
        nb_inputs: 128,
        nb_outputs: 384,
    }),
    LayerSpec::Linear(LinearSpec {
        name: "lace_fnet_gru_recurrent",
        bias: Some("lace_fnet_gru_recurrent_bias"),
        subias: Some("lace_fnet_gru_recurrent_subias"),
        weights: Some("lace_fnet_gru_recurrent_weights_int8"),
        float_weights: Some("lace_fnet_gru_recurrent_weights_float"),
        weights_idx: None,
        diag: None,
        scale: Some("lace_fnet_gru_recurrent_scale"),
        nb_inputs: 128,
        nb_outputs: 384,
    }),
    LayerSpec::Linear(LinearSpec {
        name: "lace_cf1_kernel",
        bias: Some("lace_cf1_kernel_bias"),
        subias: Some("lace_cf1_kernel_subias"),
        weights: Some("lace_cf1_kernel_weights_int8"),
        float_weights: Some("lace_cf1_kernel_weights_float"),
        weights_idx: None,
        diag: None,
        scale: Some("lace_cf1_kernel_scale"),
        nb_inputs: 128,
        nb_outputs: 16,
    }),
    LayerSpec::Linear(LinearSpec {
        name: "lace_cf1_gain",
        bias: Some("lace_cf1_gain_bias"),
        subias: None,
        weights: None,
        float_weights: Some("lace_cf1_gain_weights_float"),
        weights_idx: None,
        diag: None,
        scale: None,
        nb_inputs: 128,
        nb_outputs: 1,
    }),
    LayerSpec::Linear(LinearSpec {
        name: "lace_cf1_global_gain",
        bias: Some("lace_cf1_global_gain_bias"),
        subias: None,
        weights: None,
        float_weights: Some("lace_cf1_global_gain_weights_float"),
        weights_idx: None,
        diag: None,
        scale: None,
        nb_inputs: 128,
        nb_outputs: 1,
    }),
    LayerSpec::Linear(LinearSpec {
        name: "lace_cf2_kernel",
        bias: Some("lace_cf2_kernel_bias"),
        subias: Some("lace_cf2_kernel_subias"),
        weights: Some("lace_cf2_kernel_weights_int8"),
        float_weights: Some("lace_cf2_kernel_weights_float"),
        weights_idx: None,
        diag: None,
        scale: Some("lace_cf2_kernel_scale"),
        nb_inputs: 128,
        nb_outputs: 16,
    }),
    LayerSpec::Linear(LinearSpec {
        name: "lace_cf2_gain",
        bias: Some("lace_cf2_gain_bias"),
        subias: None,
        weights: None,
        float_weights: Some("lace_cf2_gain_weights_float"),
        weights_idx: None,
        diag: None,
        scale: None,
        nb_inputs: 128,
        nb_outputs: 1,
    }),
    LayerSpec::Linear(LinearSpec {
        name: "lace_cf2_global_gain",
        bias: Some("lace_cf2_global_gain_bias"),
        subias: None,
        weights: None,
        float_weights: Some("lace_cf2_global_gain_weights_float"),
        weights_idx: None,
        diag: None,
        scale: None,
        nb_inputs: 128,
        nb_outputs: 1,
    }),
    LayerSpec::Linear(LinearSpec {
        name: "lace_af1_kernel",
        bias: Some("lace_af1_kernel_bias"),
        subias: Some("lace_af1_kernel_subias"),
        weights: Some("lace_af1_kernel_weights_int8"),
        float_weights: Some("lace_af1_kernel_weights_float"),
        weights_idx: None,
        diag: None,
        scale: Some("lace_af1_kernel_scale"),
        nb_inputs: 128,
        nb_outputs: 16,
    }),
    LayerSpec::Linear(LinearSpec {
        name: "lace_af1_gain",
        bias: Some("lace_af1_gain_bias"),
        subias: None,
        weights: None,
        float_weights: Some("lace_af1_gain_weights_float"),
        weights_idx: None,
        diag: None,
        scale: None,
        nb_inputs: 128,
        nb_outputs: 1,
    }),
];

/// libopus `LACELayers` (`dnn/lace_data.h`): the layers of the model, in the
/// order of the C struct. Each layer borrows its arrays from a weight list.
#[derive(Clone, Copy, Default)]
pub struct LaceLayers<'w> {
    /// `lace_pitch_embedding`: float weights, 301 inputs, 64 outputs.
    pub lace_pitch_embedding: LinearLayer<'w>,
    /// `lace_fnet_conv1`: float weights, 173 inputs, 96 outputs.
    pub lace_fnet_conv1: LinearLayer<'w>,
    /// `lace_fnet_conv2`: int8 weights, 768 inputs, 128 outputs.
    pub lace_fnet_conv2: LinearLayer<'w>,
    /// `lace_fnet_tconv`: int8 weights, 128 inputs, 512 outputs.
    pub lace_fnet_tconv: LinearLayer<'w>,
    /// `lace_fnet_gru_input`: int8 weights, 128 inputs, 384 outputs.
    pub lace_fnet_gru_input: LinearLayer<'w>,
    /// `lace_fnet_gru_recurrent`: int8 weights, 128 inputs, 384 outputs.
    pub lace_fnet_gru_recurrent: LinearLayer<'w>,
    /// `lace_cf1_kernel`: int8 weights, 128 inputs, 16 outputs.
    pub lace_cf1_kernel: LinearLayer<'w>,
    /// `lace_cf1_gain`: float weights, 128 inputs, 1 output.
    pub lace_cf1_gain: LinearLayer<'w>,
    /// `lace_cf1_global_gain`: float weights, 128 inputs, 1 output.
    pub lace_cf1_global_gain: LinearLayer<'w>,
    /// `lace_cf2_kernel`: int8 weights, 128 inputs, 16 outputs.
    pub lace_cf2_kernel: LinearLayer<'w>,
    /// `lace_cf2_gain`: float weights, 128 inputs, 1 output.
    pub lace_cf2_gain: LinearLayer<'w>,
    /// `lace_cf2_global_gain`: float weights, 128 inputs, 1 output.
    pub lace_cf2_global_gain: LinearLayer<'w>,
    /// `lace_af1_kernel`: int8 weights, 128 inputs, 16 outputs.
    pub lace_af1_kernel: LinearLayer<'w>,
    /// `lace_af1_gain`: float weights, 128 inputs, 1 output.
    pub lace_af1_gain: LinearLayer<'w>,
}

impl<'w> LaceLayers<'w> {
    /// libopus `init_lacelayers`: builds every layer from the arrays of `list`,
    /// in the order of `LAYERS`. Fails at the first layer whose arrays are
    /// missing, have the wrong size or are malformed; then nothing is built.
    pub fn init(list: &WeightList<'w>) -> Result<Self, Error> {
        Ok(LaceLayers {
            lace_pitch_embedding: LAYERS[0].linear(list)?,
            lace_fnet_conv1: LAYERS[1].linear(list)?,
            lace_fnet_conv2: LAYERS[2].linear(list)?,
            lace_fnet_tconv: LAYERS[3].linear(list)?,
            lace_fnet_gru_input: LAYERS[4].linear(list)?,
            lace_fnet_gru_recurrent: LAYERS[5].linear(list)?,
            lace_cf1_kernel: LAYERS[6].linear(list)?,
            lace_cf1_gain: LAYERS[7].linear(list)?,
            lace_cf1_global_gain: LAYERS[8].linear(list)?,
            lace_cf2_kernel: LAYERS[9].linear(list)?,
            lace_cf2_gain: LAYERS[10].linear(list)?,
            lace_cf2_global_gain: LAYERS[11].linear(list)?,
            lace_af1_kernel: LAYERS[12].linear(list)?,
            lace_af1_gain: LAYERS[13].linear(list)?,
        })
    }

    /// Every layer, in the order of `LAYERS`.
    pub fn layers(&self) -> [Layer<'_, 'w>; 14] {
        [
            Layer::Linear(&self.lace_pitch_embedding),
            Layer::Linear(&self.lace_fnet_conv1),
            Layer::Linear(&self.lace_fnet_conv2),
            Layer::Linear(&self.lace_fnet_tconv),
            Layer::Linear(&self.lace_fnet_gru_input),
            Layer::Linear(&self.lace_fnet_gru_recurrent),
            Layer::Linear(&self.lace_cf1_kernel),
            Layer::Linear(&self.lace_cf1_gain),
            Layer::Linear(&self.lace_cf1_global_gain),
            Layer::Linear(&self.lace_cf2_kernel),
            Layer::Linear(&self.lace_cf2_gain),
            Layer::Linear(&self.lace_cf2_global_gain),
            Layer::Linear(&self.lace_af1_kernel),
            Layer::Linear(&self.lace_af1_gain),
        ]
    }
}