DannaSep<T>

Models & Types · AiDotNet.Audio.SourceSeparation

Danna-Sep music source separation model using dual-path attention networks.

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

Implements: IMusicSourceSeparator<T>

For Beginners

Danna-Sep takes a mixed song and separates it into individual instruments (vocals, drums, bass, other). It works by analyzing the audio from two perspectives - time patterns and frequency patterns - to figure out which parts of the mix belong to which instrument.

Usage:

How It Works

Danna-Sep (2024) uses a dual-path attention network with a novel aggregation strategy for music source separation. It processes audio in both time and frequency dimensions using interleaved attention blocks, achieving competitive SDR scores with efficient computation on the MUSDB18 benchmark.

Constructors

Constructor Summary
DannaSep(NeuralNetworkArchitecture<T> architecture, DannaSepOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer) Creates a Danna-Sep model in native training mode.
DannaSep(NeuralNetworkArchitecture<T> architecture, string modelPath, DannaSepOptions options) Creates a Danna-Sep model in native training mode.

Properties

Property Summary
IReadOnlyList<string> SupportedSources { get; }
int NumStems { get; }

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()
SourceSeparationResult<T> Separate(Tensor<T> audio)
Task<SourceSeparationResult<T>> SeparateAsync(Tensor<T> audio, CancellationToken ct)
Tensor<T> ExtractSource(Tensor<T> audio, string source)
Tensor<T> GetSourceMask(Tensor<T> audio, string source)
Tensor<T> Remix(SourceSeparationResult<T> separationResult, IReadOnlyDictionary<string, double> sourceVolumes)
Tensor<T> RemoveSource(Tensor<T> audio, string source)