Conv1d with some builtin handling of asymmetric or causal padding and normalization.

Positionwise feed forward layer. FeedForward are appied on each position of the sequence. The output dim is same with the input dim. Arg

Wrapper around Conv2d and normalization applied to this conv to provide a uniform interface across normalization approaches.

Multi-Period Discriminator module adapted from https://github.com/jik876/hifi-gan. Additionally, it allows incorporating conditional informat

Adaptive Layer Normalization module with learnable embeddings per `num_embeddings` classes Args: num_embeddings (int): Number of emb

Simple bit packer to handle ints with a non standard width, e.g. 10 bits. Note that for some bandwidth (1.5, 3), the codebook representation w

BitUnpacker does the opposite of `BitPacker`. Args: bits (int): number of bits of the values to decode. fo (IO[bytes]): file-obje

EnCodec model operating on the raw waveform. Args: target_bandwidths (list of float): Target bandwidths. encoder (nn.Module): Enco

Inverse Modified Discrete Cosine Transform (IMDCT) module. Args: frame_len (int): Length of the MDCT frame. padding (str, op

Residual block module in HiFiGAN/BigVGAN.

SEANet decoder. Args: channels (int): Audio channels. dimension (int): Intermediate representation dimension. n_filters (i

SEANet encoder. Args: channels (int): Audio channels. dimension (int): Intermediate representation dimension. n_filters (i

Residual block from SEANet model. Args: dim (int): Dimension of the input/output kernel_sizes (list): List of kernel sizes for the

LSTM without worrying about the hidden state, nor the layout of the data. Expects input as convolutional layout.

Implementation of a sine-based periodic activation function Shape: - Input: (B, C, T) - Output: (B, C, T), same shape as the

Base class of Transfomer decoder module. Args: vocab_size: output dim encoder_output_size: dimension of attention attentio

Vector quantization implementation. Currently supports only euclidean distance. Args: dim (int): Dimension codebook_size (int)

ArithmeticCoder, Let us take a distribution `p` over `N` symbols, and assume we have a stream of random variables `s_t` sampled from `p`. Let

ArithmeticDecoder, see `ArithmeticCoder` for a detailed explanation. Note that this must be called with **exactly** the same parameters and seque

Encoder layer module. Args: size (int): Input dimension. self_attn (torch.nn.Module): Self-attention module instance.

The ConstantLR scheduler This scheduler keeps a constant lr

Convolution-friendly LayerNorm that moves channels to last dimensions before running the normalization and moves them back to original positi

ConvNeXt Block adapted from https://github.com/facebookresearch/ConvNeXt to 1D audio signal. Args: dim (int): Number of input channels.

ConvolutionModule in Conformer model.

Computes a discriminator loss, given a discriminator on generated waveforms/spectrograms compared to ground truth waveforms/spectrograms.

Single decoder layer module. Args: size (int): Input dimension. self_attn (torch.nn.Module): Self-attention module instance.

Discriminator Loss module. Calculates the loss for the discriminator based on real and generated outputs.

STFT sub-discriminator. Args: filters (int): Number of filters in convolutions in_channels (int): Number of input channels. Defaul

Codebook with Euclidean distance. Args: dim (int): Dimension. codebook_size (int): Codebook size. kmeans_init (bool): Whet

Feature Matching Loss module. Calculates the feature matching loss between feature maps of the sub-discriminators.

Generator Loss module. Calculates the loss for the generator based on discriminator outputs.

Custom implementation of ISTFT since torch.istft doesn't allow custom padding (other than `center=True`) with windowing. This is because the

Language Model to estimate probabilities of each codebook entry. We predict all codebooks in parallel for a given time step. Args: n_

Label-smoothing loss. In a standard CE loss, the label's data distribution is: [0,1,2] -> [ [1.0, 0.0, 0.0], [0.0, 1.0, 0

Residual vector quantization implementation. Follows Algorithm 1. in https://arxiv.org/pdf/2107.03312.pdf

L1 distance between the mel-scaled magnitude spectrograms of the ground truth sample and the generated sample

Multi-Scale STFT (MS-STFT) discriminator. Args: filters (int): Number of filters in convolutions in_channels (int): Number of inpu

Wrapper around Conv1d and normalization applied to this conv to provide a uniform interface across normalization approaches.

Wrapper around ConvTranspose1d and normalization applied to this conv to provide a uniform interface across normalization approaches.

ResBlock adapted from HiFi-GAN V1 (https://github.com/jik876/hifi-gan) with dilated 1D convolutions, but without upsampling layers. Args

Residual Vector Quantizer. Args: dimension (int): Dimension of the codebooks. n_q (int): Number of residual vector quantizers used

ConvTranspose1d with some builtin handling of asymmetric or causal padding and normalization.

Definition of sine generator SineGen(samp_rate, harmonic_num = 0, sine_amp = 0.1, noise_std = 0.003, voiced_threshold = 0

SourceModule for hn-nsf SourceModule(sampling_rate, harmonic_num=0, sine_amp=0.1, add_noise_std=0.003, voiced_threshod=0) sa

Construct an Swish object.

Encoder layer module. Args: size (int): Input dimension. self_attn (torch.nn.Module): Self-attention module instance.

The WarmupLR scheduler This scheduler is almost same as NoamLR Scheduler except for following difference: NoamLR: lr = optimizer

From version 1.8.0, return_complex must always be given explicitly for real inputs and return_complex=False has been deprecated. De

Base class for the generator's backbone. It preserves the same temporal resolution across all layers.

Base class of Transfomer decoder module. Args: vocab_size: output dim encoder_output_size: dimension of attention attentio

Conformer encoder module.

Convolutional 1D subsampling (to 1/2 length). It is designed for Whisper, ref: https://github.com/openai/whisper/blob/main/whisper/model

Convolutional 2D subsampling (to 1/4 length). Args: idim (int): Input dimension. odim (int): Output dimension. dropout_ra

Convolutional 2D subsampling (to 1/6 length). Args: idim (int): Input dimension. odim (int): Output dimension. dropout_rat

Convolutional 2D subsampling (to 1/8 length). Args: idim (int): Input dimension. odim (int): Output dimension. dropout_ra