Implements exponential moving average shadowing for your model. Utilizes an inverse decay schedule to manage longer term training runs.

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

Base implementation of a pattern over a sequence with multiple codebooks. The codebook pattern consists in a layout, defining for each sequence s

Accumulates log values over N steps and flushes averaged metrics. Avoids per-step CUDA syncs (.item()) and prevents PL from caching stale val

Adaptive log-SNR distribution shift. Maps t∈[0,1] to log-SNR-spaced values while preserving order (0→0, 1→1). Equivalent to applying: logsnr

Pseudo Quadrature Mirror Filter (PQMF) for multiband signal decomposition and reconstruction. Uses polyphase representation which is computat

Psychoacoustic noise generator that works directly in frequency domain to avoid filterbank reconstruction artifacts

A noise sampler backed by a torchsde.BrownianTree. Args: x (Tensor): The tensor whose shape, device and dtype to use to generate

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

DPM-Solver. See https://arxiv.org/abs/2206.00927.

Flux/SD3/Self-Flow timestep shift: t_shifted = alpha * t / (1 + (alpha-1) * t). Convention: t=0 is data, t=1 is noise. alpha > 1 shifts times

No-op distribution shift — returns timesteps unchanged.

A module that applies multiple conditioners to an input dictionary based on the keys Args: conditioners: a dictionary of conditioner

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

Randomly invert the phase of a signal

STFT magnitude loss module. See [Arik et al., 2018](https://arxiv.org/abs/1808.06719) and [Engel et al., 2020](https://arxiv.org/abs/2001.046

UNet1d with Classifier-Free Guidance

Volume normalization and augmentation of a signal [LUFS standard]

Wrapper for ARC post-training on a conditional audio diffusion model.

FAD metrics for autoencoder training. Compares reconstructed reference audio vs original reference audio using Frechet Audio Distance. Uses t

A wrapper around torchsde.BrownianTree that enables batches of entropy.

Implements Adam algorithm with weight decay fix as introduced in [Decoupled Weight Decay Regularization](https://arxiv.org/abs/1711.05101).

Feature matching loss using CLAP (HTSAT) intermediate layer features. Extracts features from the HTSAT Swin Transformer BasicLayer stages in

Wrapper for training a conditional audio diffusion model.

PANNs CNN14 model for audio embeddings.