CMGANOptions

Options & Configuration · AiDotNet.Audio.Enhancement

Configuration options for the CMGAN (Conformer-based Metric GAN) speech enhancement model.

Inheritance: ModelOptions → CMGANOptions

For Beginners

CMGAN uses a “competition” strategy to enhance speech:

  • A “generator” network tries to clean up noisy audio
  • A “discriminator” network judges how realistic the cleaned audio sounds
  • They train together: the generator gets better at fooling the discriminator
  • The result is very natural-sounding enhanced speech

The “Conformer” part means it uses a mix of attention (looking at the whole signal) and convolution (looking at local patterns), getting the best of both approaches.

How It Works

CMGAN (Cao et al., INTERSPEECH 2022) combines a conformer-based generator with a metric discriminator for high-quality speech enhancement. It achieves PESQ 3.41 and STOI 0.97 on the VoiceBank-DEMAND dataset, outperforming previous GAN-based methods.

References:

  • Paper: “CMGAN: Conformer-based Metric GAN for Speech Enhancement” (Cao et al., INTERSPEECH 2022)
  • Repository: https://github.com/ruizhecao96/CMGAN

Constructors

Constructor Summary
CMGANOptions()

Properties

Property Summary
OnnxModelOptions OnnxOptions { get; set; } Gets or sets ONNX runtime options.
bool EnhancePhase { get; set; } Gets or sets whether to enhance both magnitude and phase.
bool UseDiscriminator { get; set; } Gets or sets whether to use the metric discriminator during training.
double DiscriminatorLearningRate { get; set; } Gets or sets the discriminator learning rate.
double DropoutRate { get; set; } Gets or sets the dropout rate.
double EnhancementStrength { get; set; } Gets or sets the enhancement strength (0.0 = no enhancement, 1.0 = maximum).
double LearningRate { get; set; } Gets or sets the generator learning rate.
double PowerLawCompressionExponent { get; set; } Gets or sets the power-law compression exponent applied to the input spectrogram.
double WeightDecay { get; set; } Gets or sets the weight decay for regularization.
int ConformerDim { get; set; } Gets or sets the Conformer hidden dimension.
int ConformerKernelSize { get; set; } Gets or sets the convolution kernel size in Conformer layers.
int DecoderKernelSize { get; set; } Gets or sets the decoder kernel size.
int FftSize { get; set; } Gets or sets the FFT window size.
int HopLength { get; set; } Gets or sets the hop length between frames.
int NumAttentionHeads { get; set; } Gets or sets the number of attention heads in Conformer layers.
int NumConformerLayers { get; set; } Gets or sets the number of Conformer encoder layers.
int NumFreqBins { get; set; } Gets or sets the number of frequency bins.
int SampleRate { get; set; } Gets or sets the audio sample rate in Hz.
int[] EncoderChannels { get; set; } Gets or sets the number of U-Net encoder channels.
string ModelPath { get; set; } Gets or sets the path to a pre-trained ONNX model file.