Implementation:Facebookresearch Audiocraft MultiPeriodDiscriminator

Knowledge Sources	Facebookresearch_Audiocraft HiFi-GAN
Domains	Audio_Synthesis, GAN
Last Updated	2026-02-14 01:00 GMT

Overview

Concrete tool for analyzing audio waveforms at different periodic intervals to capture fine-grained harmonic patterns for adversarial training.

Description

MultiPeriodDiscriminator implements the MPD from HiFi-GAN. It reshapes 1D audio waveforms into 2D representations using different period values (default [2,3,5,7,11]) and applies 2D convolutions to each. This design captures periodic patterns at different fundamental frequencies, enabling the discriminator to assess both harmonic quality and temporal structure.

Usage

Import this class when building adversarial training pipelines for audio compression or generation models that need a period-based discriminator for quality assessment.

Code Reference

Source Location

• Repository: Facebookresearch_Audiocraft
• File: audiocraft/adversarial/discriminators/mpd.py
• Lines: 1-106

Signature

class MultiPeriodDiscriminator(MultiDiscriminator):
    def __init__(self, in_channels: int = 1, out_channels: int = 1,
                 n_layers: int = 5, periods: tp.Optional[tp.List[int]] = None):
        """
        Args:
            in_channels: Number of input audio channels.
            out_channels: Output channels per sub-discriminator.
            n_layers: Number of convolution layers per period.
            periods: List of period values (default [2,3,5,7,11]).
        """

Import

from audiocraft.adversarial.discriminators.mpd import MultiPeriodDiscriminator

I/O Contract

Inputs

Name	Type	Required	Description
x	torch.Tensor	Yes	Audio waveform [B, C, T]

Outputs

Name	Type	Description
logits	list[torch.Tensor]	Discriminator logits from each period sub-discriminator
features	list[list[torch.Tensor]]	Intermediate features for feature matching loss

Usage Examples

from audiocraft.adversarial.discriminators.mpd import MultiPeriodDiscriminator
import torch

mpd = MultiPeriodDiscriminator(in_channels=1)
wav = torch.randn(4, 1, 24000)
logits, features = mpd(wav)

Related Pages

• Principle:Facebookresearch_Audiocraft_Multi_Period_Discrimination