espnet2.enh.diffusion.sampling.predictors.Predictor

source

class espnet2.enh.diffusion.sampling.predictors.Predictor(sde, score_fn, probability_flow=False)

Bases: ABC

The abstract class for a predictor algorithm.

debug_update_fn(x, t, *args)

abstract update_fn(x, t, *args)

One update of the predictor.

• Parameters:
  • x – A PyTorch tensor representing the current state
  • t – A Pytorch tensor representing the current time step.
  • *args – Possibly additional arguments, in particular y for OU processes
• Returns: A PyTorch tensor of the next state. x_mean: A PyTorch tensor. The next state without random noise.

  Useful for denoising.

• Return type: x