AudioSeparator.Onnx.Demucs 0.2.0-beta.2

AudioSeparator.Onnx.Demucs

Demucs/htdemucs ONNX backend for AudioSeparator — separate audio into drums, bass, vocals, and other stems.

Part of AudioSeparator — modular .NET audio stem separation.

Install

dotnet add package AudioSeparator.Onnx.Demucs
dotnet add package AudioSeparator.NAudio

Or swap NAudio for FFMPEG when you need broader format support:

dotnet add package AudioSeparator.FFMPEG

When to use this package

• You want to separate audio stems using a Demucs or htdemucs ONNX model.
• This is the recommended entry point for most applications.
• Pair with NAudio (simple WAV/MP3) or FFMPEG (any format ffmpeg supports).

Core principle

The separator returns SeparationResult with separated stems. It does not write files. Call WriteToDirectoryAsync after RunAsync to save stems to disk.

Defaults

DemucsBuilder.Create(...) pre-configures htdemucs requirements:

Override with .UseStemNames(...) or .WithRequirements(...) when needed.

Quick start

using AudioSeparator.NAudio;
using AudioSeparator.Onnx.Demucs;
using Microsoft.ML.OnnxRuntime;

var builder = DemucsBuilder.Create("htdemucs.onnx")
    .UseNAudio()
    .ConfigureSessionOptions(options =>
    {
        options.GraphOptimizationLevel = GraphOptimizationLevel.ORT_ENABLE_ALL;
        options.AppendExecutionProvider_CUDA(0); // optional GPU
    });

using var separator = builder.Build();
var session = await separator.CreateSession("input.wav");

foreach (var task in session.Tasks)
{
    task.SetProgressCallback((current, total) =>
        Console.WriteLine($"{task.Description}: {current}/{total}"));
}

var result = await session.RunAsync();
await result.WriteToDirectoryAsync("./Outputs");

foreach (var (name, _) in result.Stems)
{
    Console.WriteLine($"Wrote stem: {name}");
}

FFMPEG output (e.g. MP3 stems)

using AudioSeparator.FFMPEG;
using AudioSeparator.Onnx.Demucs;

var builder = DemucsBuilder.Create("htdemucs.onnx")
    .UseFFMPEG(options =>
    {
        options.OutputFormat = "mp3";
        options.OutputCodec = "libmp3lame";
    });

Session flow

Key API

Inherited from OnnxSeparatorBuilder: .UseAudio(), .UseStemNames(), .WithRequirements(), .ConfigureSessionOptions().

Requirements

• .NET 10
• ONNX model file (e.g. htdemucs) — not included in this package
• AudioSeparator.Onnx (transitive)
• CUDA (optional) — GPU inference via ConfigureSessionOptions
• ffmpeg/ffprobe (optional) — only if using FFMPEG extensions

Examples

Sample projects in the AudioSeparator repository:

• Examples/AudioSeparator.Console — minimal console usage
• Examples/AudioSeparator.Spectre — progress bars with Spectre.Console
• Examples/AudioSeparator.Benchmark — performance benchmarking

Related packages

Source: AudioSeparator.Onnx.Demucs on GitHub

License

MIT — see AudioSeparator repository.