Shiny.Audio.Linux 3.0.0-beta-0035

Shiny.Audio

Cross-platform audio capture and playback for .NET MAUI, iOS, Mac Catalyst, macOS, Android, Windows, Browser (WebAssembly), and Linux (via the companion Shiny.Audio.Linux package).

Shiny.Audio provides the low-level audio primitives that back Shiny.Speech, but is usable on its own:

• IAudioSource — microphone capture that streams raw PCM audio (16kHz, 16-bit, mono) via a thread-safe PipeStream, with runtime permission handling (RequestAccess / AccessState) and a normalized InputLevelChanged VU signal on every platform.
• IAudioPlayer — stream / file / URL playback (e.g. MP3) with optional normalized AudioLevelChanged metering for VU-style UI. Clips play concurrently: StartAsync returns an IAudioPlayback you can stop on its own, PlayAsync waits for one clip without interrupting the rest, and StopAsync stops the lot.
• IAudioRecorder — record the microphone straight to a WAV file, optionally through a live effect chain, capturing the processed take, the raw one, or both.
• IAudioDuplex — full-duplex voice for calls: echo-cancelled mic frames out (Captured) and live network PCM in (Play) on one voice-processing session, with a jitter buffer (priming, underrun re-prime, overflow drop) and live speakerphone switching. iOS, Mac Catalyst, macOS and Android.
• AudioEffectChain — real-time DSP on capture (pitch shift, echo, reverb, filters, distortion, ring modulation, chorus, gain, noise gate) that you toggle and re-tune while recording.
• AudioLevel — the shared dBFS mapping behind every meter (FromRms / FromPcm16 / FromSamples), so input and output bars read on the same scale — including PCM you meter yourself.
• WavWriter / WavReader — streaming RIFF/WAVE PCM I/O, plus AudioEffectProcessor for applying an effect chain to a file that already exists.

Getting Started

using Shiny;

builder.Services.AddAudioServices(); // registers IAudioSource, IAudioPlayer, IAudioRecorder, ...

using Shiny.Audio;

public class Recorder(IAudioSource audioSource)
{
    public async Task RecordAsync(CancellationToken ct)
    {
        if (await audioSource.RequestAccess() != AccessState.Available)
            return;

        var pcmStream = await audioSource.StartCaptureAsync(cancellationToken: ct);
        // consume pcmStream ...
        await audioSource.StopCaptureAsync();
    }
}

Full-duplex voice (calls)

var session = await audio.Duplex.StartAsync(new AudioDuplexOptions { SampleRate = 16000 });

// mic -> network: one 20 ms PCM16 mono frame per item, echo-cancelled
await foreach (var frame in session.Captured.ReadAllAsync(ct))
    await network.SendAsync(frame, ct);

// network -> speaker, from any thread, any size (jitter-buffered; gaps play as silence)
session.Play(received.Span);

session.Speakerphone = false;   // earpiece, live
await session.StopAsync();      // restores the audio session

Defaults: 16 kHz, AudioProcessingOptions.VoiceChat, speakerphone on, 20 ms frames, 60 ms target playback latency and 200 ms maximum (older audio is dropped). One session at a time. Not available on Windows, Linux or the browser. On Android add MODIFY_AUDIO_SETTINGS to the manifest.

Effects & recording

Effects are caller-owned objects. Build a chain, hand it to a capture or recording session, and keep the reference — toggling an effect or moving a parameter applies on the next audio buffer, with no restart and no clicks.

using Shiny.Audio;

var chain = new AudioEffectChain();
var pitch = chain.Add(new PitchShiftEffect { Semitones = 0 });
var echo  = chain.Add(new EchoEffect { Enabled = false });

await recorder.StartAsync(new AudioRecordingOptions
{
    Mode = AudioRecordMode.Both,   // writes the processed and the raw take
    Effects = chain
});

pitch.Semitones = 5;      // heard immediately
echo.Enabled = true;
chain.Enabled = false;    // master bypass

var recording = await recorder.StopAsync();

Available effects: GainEffect, NoiseGateEffect, BiquadFilterEffect, DistortionEffect, RingModEffect, EchoEffect, ChorusEffect, ReverbEffect, PitchShiftEffect — plus AudioEffectPresets.Create(...) for ready-made combinations (Robot, Chipmunk, DeepVoice, Cathedral, Telephone, Megaphone, Ensemble).

Choosing the microphone

IAudioDevices lists the microphones the system sees (GetInputs(), with CurrentInput the system default) and raises Changed when one is plugged in, removed or made the default. Hand one to a recording or a capture session; null keeps the system default.

var mic = devices.GetInputs().FirstOrDefault(x => x.Type == AudioDeviceType.Usb);

await recorder.StartAsync(new AudioRecordingOptions { InputDevice = mic });
var pcm = await audioSource.StartCaptureAsync(new AudioCaptureOptions { InputDevice = mic });

A microphone unplugged since it was listed falls back to the system default rather than failing.

Do not put effects on audio bound for speech recognition — they destroy recognition and wake-word accuracy.

Linux

Linux ships as a separate Shiny.Audio.Linux package (it carries a managed MP3 decoder, since Linux has no system decoder to call). Add it and register before anything else:

builder.Services.AddLinuxAudio();   // no-op when not running on Linux

It supplies all four services — IAudioSource, IAudioPlayer, IAudioDevices and IAudioMonitor — over PulseAudio/PipeWire, falling back to ALSA on headless systems and containers.

See https://shinylib.net/speech for full documentation.