DCCRN<T>
Models & Types · AiDotNet.Audio.Enhancement
DCCRN - Deep Complex Convolution Recurrent Network for speech enhancement.
Inheritance: NeuralNetworkBase<T> → AudioNeuralNetworkBase<T> → DCCRN<T>
Implements: IAudioEnhancer<T>
For Beginners
DCCRN is a neural network designed specifically for cleaning up noisy audio. Unlike simpler methods that only work with the “loudness” of frequencies, DCCRN also considers the “timing” (phase), which results in more natural-sounding enhanced audio.
Think of it like this: regular enhancement is like adjusting volume of different frequencies, while DCCRN can also adjust the timing of sound waves to better reconstruct the original clean speech.
Usage:
How It Works
DCCRN operates directly on complex-valued spectrograms, preserving phase information for high-quality speech enhancement. Key features:
- Complex-valued convolutions for better spectral modeling
- LSTM layers for temporal dependencies
- Skip connections for gradient flow
- Mask-based enhancement for clean speech estimation
Reference: “DCCRN: Deep Complex Convolution Recurrent Network for Phase-Aware Speech Enhancement” by Hu et al., Interspeech 2020
Constructors
| Constructor | Summary |
|---|---|
DCCRN(NeuralNetworkArchitecture<T> architecture, DCCRNOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction) |
Creates a DCCRN model for native training and inference. |
DCCRN(NeuralNetworkArchitecture<T> architecture, string modelPath, DCCRNOptions options, OnnxModelOptions onnxOptions) |
Creates a DCCRN model for native training and inference. |
Properties
| Property | Summary |
|---|---|
bool SupportsTraining { get; } |
Gets whether this network supports training. |
bool UseComplexMask { get; } |
Gets whether complex mask is used. |
double EnhancementStrength { get; set; } |
|
int LatencySamples { get; } |
|
int NumChannels { get; set; } |
|
int NumStages { get; } |
Gets the number of encoder/decoder stages. |
Methods
| Method | Summary |
|---|---|
Dictionary<string, Tensor<T>> GetNamedLayerActivations(Tensor<T> input) |
Returns per-layer activations. |
ModelMetadata<T> GetModelMetadata() |
|
Tensor<T> Enhance(Tensor<T> audio) |
|
Tensor<T> EnhanceWithReference(Tensor<T> audio, Tensor<T> reference) |
|
Tensor<T> ForwardForTraining(Tensor<T> input) |
Training forward pass: consumes a complex STFT [B, 2, F, T] and returns the enhanced STFT. |
Tensor<T> ProcessChunk(Tensor<T> audioChunk) |
|
void EstimateNoiseProfile(Tensor<T> noiseOnlyAudio) |
|
void ResetState() |