AudioSuperResolution<T>

Models & Types · AiDotNet.Audio.Effects

Audio Super-Resolution model for upsampling low-resolution audio to high-resolution (Kuleshov et al., 2017; Li et al., 2021).

Inheritance: NeuralNetworkBase<T> → AudioNeuralNetworkBase<T> → AudioSuperResolution<T>

Implements: IAudioEnhancer<T>

For Beginners

Audio Super-Resolution is like AI-powered upscaling for sound. Just as image super-resolution makes blurry photos sharper, this model makes low-quality audio sound clearer and more detailed.

Common uses:

  • Upscaling old telephone recordings (8 kHz to 44.1 kHz)
  • Recovering quality from heavily compressed audio (MP3 at 64 kbps)
  • Enhancing voice recordings from cheap microphones
  • Restoring bandwidth-limited historical recordings

How it works:

  1. Takes a low-resolution audio waveform as input
  2. Passes through residual blocks that learn to predict missing high-frequency content
  3. Outputs a high-resolution waveform with restored detail

Usage:

How It Works

Audio Super-Resolution uses deep neural networks to predict missing high-frequency content in low-resolution audio. Given input at a low sample rate (e.g., 8 kHz telephone quality), it reconstructs audio at a higher sample rate (e.g., 44.1 kHz studio quality) by predicting the missing frequency bands. The architecture uses residual blocks with attention modules to capture both local and global spectral patterns.

Constructors

Constructor Summary
AudioSuperResolution(NeuralNetworkArchitecture<T> architecture, AudioSuperResolutionOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer) Creates an Audio Super-Resolution model in native training mode.
AudioSuperResolution(NeuralNetworkArchitecture<T> architecture, string modelPath, AudioSuperResolutionOptions options) Creates an Audio Super-Resolution model in native training mode.

Properties

Property Summary
double EnhancementStrength { get; set; }
int LatencySamples { get; }
int NumChannels { get; }

Methods

Method Summary
Dictionary<string, Tensor<T>> GetNamedLayerActivations(Tensor<T> input) Collects per-block activations along the actual U-Net dataflow.
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 — the same tape-aware U-Net path as inference.
Tensor<T> ProcessChunk(Tensor<T> audioChunk)
void EstimateNoiseProfile(Tensor<T> noiseOnlyAudio)