Linux

Linux support ships in a separate Shiny.Audio.Linux package. It implements capture, playback, devices and the live monitor over PulseAudio/PipeWire, falling back to ALSA where no sound server is running — headless servers, minimal containers, Raspberry Pi images.

It is a separate package because it carries a managed MP3 decoder (NLayer): Linux has no system media decoder to hand a stream to, unlike MediaPlayer on Windows/Android or AVAudioPlayer on Apple.

dotnet add package Shiny.Audio.Linux

using Shiny;

// Registers IAudioSource + IAudioPlayer + IAudioDevices + IAudioMonitor + IAudio.
// A no-op when not running on Linux, so it is safe in shared startup code.
builder.Services.AddLinuxAudio();

Recording

AddLinuxAudio() doesn’t register IAudioRecorder. The recorder is platform-agnostic and runs on top of the Linux IAudioSource, so add AddAudioRecorder() (or AddAudioServices()) after AddLinuxAudio() if you record. Otherwise IAudio.Recorder throws. Effects work unchanged.

builder.Services.AddLinuxAudio();
builder.Services.AddAudioServices();   // adds IAudioRecorder; the Linux registrations above win

Call it first

AddLinuxAudio() must come before AddAudioServices(), AddSpeechServices() or any AddCloudSpeechToText / AddCloudTextToSpeech overload. Those all use TryAdd, so the first registration wins — and on Linux they register nothing at all.

There is no native speech on Linux

Linux has no OS speech engine to wrap the way iOS, Android and Windows do, so AddSpeechToText() and AddTextToSpeech() register nothing there. Pair AddLinuxAudio() with any cloud provider — they are pure HTTP over the PCM stream and work unchanged:

builder.Services.AddLinuxAudio();
builder.Services.AddAzureSpeech("subscription-key", "region");

Behaviour notes

• No resampler ships in the library. Streams open at exactly the format required — 16 kHz mono for capture, a decoded clip’s own rate for playback. PulseAudio converts server-side; ALSA’s default device converts through its plug plugin.

• Playback decodes MP3 and WAV only. That covers what Azure, OpenAI and ElevenLabs return by default. Any other container throws NotSupportedException naming the format.

• Playback metering works here (unlike Windows/Browser) because the player decodes to PCM in managed code and sees every sample.

• Voice processing is best-effort. PulseAudio publishes echo cancellation as a virtual source from module-echo-cancel; when that module is loaded, AudioProcessingOptions.EchoCancellation selects it, otherwise capture is raw.

• No permission prompt. Microphone access on Linux is a filesystem/group concern, so RequestAccess() returns Available whenever a backend is present.

• ShowOutputPicker() is a no-op — there is no system route picker equivalent to iOS’s AVRoutePickerView. Routing lives in the desktop environment’s sound settings. IAudioDevices.Changed fires on PulseAudio/PipeWire only, since ALSA has no change notification.

• No full-duplex call audio. IAudioDuplex isn’t implemented on Linux.

• The live monitor runs in managed code. PulseAudio’s simple API and ALSA have no input-to-output routing, so the monitor copies a capture stream into a playback stream itself. Latency is the two stream buffers added together. That’s fine for talkback or a megaphone, but not for live performance monitoring.

Runtime dependencies

The distro’s own audio libraries — libpulse-simple.so.0 (usually the libpulse0 package) or libasound.so.2 (usually libasound2). In a container, install one of them explicitly.

Probe at startup when the app must degrade gracefully rather than throw on first use:

if (!LinuxAudioServiceCollectionExtensions.IsLinuxAudioAvailable)
    logger.LogWarning("No Linux audio backend found — voice features disabled");