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

A modified Snake function which uses separate parameters for the magnitude of the periodic components Shape: - Input: (B, C, T)

Multi-headed attention from 'Attention Is All You Need' paper

BigVGAN is a neural vocoder model that applies anti-aliased periodic activation for residual blocks (resblocks). New in BigVGAN-v2: it can op

A DiT block with adaptive layer norm zero (adaLN-Zero) conditioning.

Diffusion model with a Transformer backbone.

The final layer of DiT.

Downloader for Librispeech dataset.

Multi-headed attention from 'Attention Is All You Need' paper

Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`QwenDecoderLayer`] Args: config: KimiAudioCon

This module produces sinusoidal positional embeddings of any length. Padding symbols are ignored.

Embeds scalar timesteps into vector representations.

Transformer decoder consisting of *config.decoder_layers* layers. Each layer is a [`WhisperDecoderLayer`] Args: config: WhisperConfi

Transformer encoder consisting of *config.encoder_layers* self attention layers. Each layer is a [`WhisperEncoderLayer`]. Args:

AMPBlock applies Snake / SnakeBeta activation functions with trainable parameters that control periodicity, defined for each layer. AMPBlock1

AMPBlock applies Snake / SnakeBeta activation functions with trainable parameters that control periodicity, defined for each layer. Unlike AM

Assumes filter size 12, replication padding on upsampling/downsampling, and logscale alpha/beta parameters as inputs. The hyperparameters are

Dataset for supervised fine-tuning.