BSRoFormer<T>

Models & Types · AiDotNet.Audio.SourceSeparation

BS-RoFormer (Band-Split Rotary Transformer) for music source separation.

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

Implements: IMusicSourceSeparator<T>

For Beginners

BS-RoFormer separates a mixed song into stems (vocals, drums, bass, other) by dividing frequencies into bands and using AI to figure out which parts belong to which instrument.

Usage:

How It Works

BS-RoFormer (Lu et al., 2023) splits the spectrogram into frequency bands, processes each with Transformer layers using rotary position embeddings, then fuses results for mask estimation. Achieves 12.8 dB SDR on vocals separation (MUSDB18-HQ).

Constructors

Constructor Summary
BSRoFormer(NeuralNetworkArchitecture<T> architecture, BSRoFormerOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)
BSRoFormer(NeuralNetworkArchitecture<T> architecture, string modelPath, BSRoFormerOptions options)

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)