Audiola

A modern cross-platform audio studio — arrange, separate, remix, master and master in 3D.

A full audio-production studio for Windows, macOS, and Linux built on .NET 10 + Avalonia, with a companion .NET 10 Blazor WebAssembly web app. Drag a song in, split it into stems, edit on a multitrack timeline, swap or synthesize voices with local AI, master to broadcast loudness and export an immersive 3D mix — all in one app that provisions everything it needs.

 

---

Table of contents

• What is Audiola?
• Feature overview
• Screenshots
• Features in depth
  • Studio — multitrack timeline
  • Stem separation
  • Track editor
  • Equalizer
  • Mastering
  • Spatial Audio (3D / immersive)
  • Voices — local & cloud TTS and voice swap
  • Transcription & lyrics
  • Provenance analysis
  • Project format, undo & export
• Requirements
• Install
• Self-provisioning & GPU / CUDA
• Build & run from source
• Web version
• Tech stack
• Releasing
• Documentation site
• License

---

What is Audiola?

Audiola is a desktop audio studio that brings the heavy parts of modern music production into one Fluent-design app: a multitrack timeline, AI stem separation, a focused waveform editor, an interactive equalizer, broadcast-grade mastering, immersive 3D / spatial-audio rendering and a full local + cloud voice toolkit (text-to-speech and voice swapping).

It is self-provisioning — the AI/ML features run through a managed Python environment that the app sets up, populates and updates for you. There is no manual pip install to run; click a button and Audiola downloads the model it needs into its own isolated environment.

Audiola and Singola ship as Windows, macOS, and Linux desktop apps and are accompanied by a Blazor WebAssembly web companion.

---

Feature overview

	Area	What it does
🎛️	Studio	Multitrack timeline: drag clips in, cut/trim, per-clip volume/pan/mute/solo, fades, snapping, zoom, per-track colors, real-time spectrum, one-click Export Mix.
🧩	Stem separation	Demucs (htdemucs / htdemucs_ft / htdemucs_6s / mdx) and high-quality separation via audio-separator (BS-RoFormer, Mel-Band RoFormer karaoke, 6-stem Demucs).
✂️	Track editor	Focused waveform editor: trim, delete, silence, fade in/out, normalize, echo, reverb, stereo widen, reverse, vocal cleanup, live FX preview, undo.
🎚️	Equalizer	Interactive 4-band EQ with a live Master-EQ on the studio mix.
🏆	Mastering	EQ → compressor → loudness (LUFS / BS.1770 / EBU R128) chain, A/B preview, genre presets, custom profiles and batch processing.
🌐	Spatial Audio	Position tracks in 3D, render binaural (HRTF) for headphones, or multichannel WAV (5.1 / 7.1 / 7.1.4 Atmos bed).
🗣️	Voices	Local AI models (Kokoro, Qwen3-TTS, XTTS v2, Chatterbox, seed-vc) + ElevenLabs. Create voices, swap a clip's voice, add a track from text.
📝	Transcription	Whisper transcription → LRC lyrics + embedded lyrics tag.
🔎	Provenance	Read-only C2PA / metadata / AI-generation marker analysis.
💾	Project & export	Native .audiola project bundle, full undo/redo with visual history, export to WAV / MP3 / AAC-M4A.

---

Screenshots

The app uses a dark Fluent (Mica) theme with a frequency-gradient accent and a real-time spectrum visualizer.

Studio — the multitrack workspace

Mastering	Interactive Equalizer
Spatial Audio (3D / immersive)	Voices — local & cloud
Swap a clip's voice	Add a track from text (TTS)

_{All screenshots live in assets/screenshots/ and are also shown on the documentation site.}

---

Features in depth

🎛️ Studio — multitrack timeline

A timeline workspace where the whole mix comes together.

• Drag audio files straight onto the timeline (WAV, MP3, FLAC, AIFF, M4A, OGG) and arrange clips across as many tracks as you like.
• Clip editing: split at the playhead, trim either edge with snapping, cut a region out, delete, and drag clips freely between tracks.
• Per-clip controls: gain (−24…+6 dB) and independent fade-in / fade-out handles drawn right on the clip.
• Per-track controls: volume, pan, mute, solo, enable, inline rename, custom accent color from an 8-color palette, and one-click duplicate track.
• Timeline tools: zoom (10–200 px/s), adjustable lane height (80–320 px), a snapping grid (0.1 / 0.25 / 0.5 / 1.0 s), a draggable playhead with auto-scroll, and a global selection range with loop playback.
• Real-time spectrum visualizer (audioMotion-style) right in the timeline header.
• Master mix: master volume plus a live Master-EQ, and a one-click Export Mix that renders every track with its effects.

🧩 Stem separation

Pull a song apart into its instruments — added straight back onto the timeline as new tracks.

• Demucs (local, via Python): htdemucs, htdemucs_ft (fine-tuned, the default), htdemucs_6s (6-stem) and the mdx_extra / mdx_extra_q models. Optional test-time augmentation (shifts) trades speed for quality.
• High-quality separation via audio-separator:
  • BS-RoFormer — state-of-the-art vocals / instrumental.
  • Mel-Band RoFormer (karaoke) — separates lead from background vocals.
  • Demucs 6-stem — vocals, drums, bass, guitar, piano, other.
• Stems extracted: vocals, drums, bass, guitar, piano and other.
• Smart / auto mode runs a 6-stem split with content detection, automatically dropping near-silent stems below a sensible threshold.

✂️ Track editor

Double-click a clip to open a focused, large waveform editor with non-destructive edits and live preview.

• Edits: trim to selection, delete (close the gap), silence a selection, linear fade-in / fade-out.
• Effects: normalize, echo (delay + feedback + mix), reverb (Schroeder reverberator), stereo widen (mid/side), reverse, and a one-click vocal cleanup (high-pass, harshness taming, air shelf, de-esser and gentle compression).
• Live FX preview for echo, reverb and stereo widen with real-time parameter changes during playback.
• Undo and per-edit export to WAV / MP3 / M4A.

🎚️ Equalizer

• Interactive 4-band EQ — low-shelf @ 100 Hz, peaking @ 500 Hz, peaking @ 3 kHz, high-shelf @ 10 kHz — with draggable curve points.
• A live Master-EQ applied to the studio mix while you listen.
• Export the EQ-processed mix or a single loaded track.

🏆 Mastering

A professional mastering chain on your studio mix.

• Chain: EQ (high-pass + low shelf + peaking mid + high shelf) → compressor (stereo-coupled, feed-forward) → loudness normalization.
• Loudness metering uses ITU-R BS.1770 / EBU R128 integrated LUFS, with K-weighting and gating.
• A/B preview between the original and mastered signal, with a live master processor.
• Genre presets — Streaming −14 LUFS, Pop, Rock, Radio-Ready, Loud/Club, Warm/Vintage, Podcast, Bass-Boost, HipHop/Trap, EDM, Acoustic/Folk, Lo-Fi, Vocal-Boost — plus your own savable custom profiles.
• Target-loudness normalization to a LUFS target of your choice.
• Batch / bulk mastering: master many files at once with an output-filename template and format conversion (e.g. convert MP3 → WAV), with per-file progress.

🌐 Spatial Audio (3D / immersive)

Place your studio tracks in 3D space and render an immersive mix.

• Position each source by azimuth, elevation, distance and level, with a top-down radar you can drag.
• Auto-arrange intelligently places common track types (vocals center, bass center, drums front, guitar/piano to the sides, etc.).
• Live preview of the spatial mix while you adjust.
• Binaural (HRTF) export for headphones — ITD/ILD, head-shadow filtering and optional room reverb.
• Multichannel WAV export: 5.1, 7.1 and 7.1.4 (Atmos bed) layouts written as WAVE_FORMAT_EXTENSIBLE with the correct channel masks.

Note: Audiola is not a licensed Dolby Atmos bitstream encoder. It produces spatial audio — binaural stereo or a 7.1.4 multichannel WAV — that you can import into an Atmos renderer or DAW for certified encoding.

🗣️ Voices — local & cloud TTS and voice swap

A complete voice toolkit, with local AI models that the app downloads and runs for you.

• Local models (managed Python environment, one-click download):
  • Kokoro — fast, CPU-friendly preset TTS.
  • Qwen3-TTS — multilingual TTS with zero-shot voice cloning.
  • Coqui XTTS v2 — multilingual TTS with zero-shot cloning.
  • Chatterbox — TTS with voice cloning from a sample.
  • seed-vc — zero-shot voice conversion (speech-to-speech) that preserves melody, timing and emphasis — it works for singing, not just speech.
• Create your own voices by recording from your microphone or uploading a sample.
• Swap the voice of a clip locally with seed-vc (keeps the performance/timing, changes only the timbre) or via ElevenLabs speech-to-speech.
• Add a new track from text (TTS) using the same voice picker, with speed and (for ElevenLabs) stability/similarity controls.
• Device selection (auto / cuda / cpu / directml) with a visible GPU check and a one-click Install CUDA Torch button.

📝 Transcription & lyrics

• Whisper transcription with selectable model sizes: tiny, base, small, medium, large-v3 and turbo.
• Produces LRC time-synced lyrics and can embed a lyrics metadata tag into exported tracks.

🔎 Provenance analysis

• A read-only analyzer that scans an audio file for C2PA content credentials, IPTC AI-generation markers, XMP/ID3 metadata and generator tags.
• Optionally reads a full C2PA manifest via c2patool if installed.
• Explains how multi-layer detection works (metadata, watermark, ML classifier). It only reports — it makes no changes to your audio.

💾 Project format, undo & export

• Native .audiola project — a single archive bundling the manifest plus all media, so clips, fades, gains, track colors, EQ bands, mastering settings, spatial layout and lyrics are all preserved (no stem re-extraction needed).
• Full undo/redo with a visual history you can click to jump to any earlier state.
• Export the full mix, a selection range or an individual track to WAV, MP3 or AAC-M4A.

---

Requirements

• Windows 10 / 11 (x64), macOS (Apple Silicon or Intel), or Linux x64.
• Running the installer: nothing else — the released build is self-contained.
• AI/ML features (stem separation, local voices, transcription): a Python interpreter on your system that Audiola can use as the base for its managed environment. Everything else (model downloads, dependency installs) is handled in-app.
• Optional NVIDIA GPU + CUDA for dramatically faster stem separation, voice conversion and transcription.
• Optional ElevenLabs API key for cloud speech-to-speech / TTS.

---

Install

Use the download page to select the recommended package, or download it from Releases:

• Windows x64: Audiola-win-x64-Setup.exe
• macOS Apple Silicon / Intel: Audiola-osx-arm64-Setup.pkg / Audiola-osx-x64-Setup.pkg
• Linux x64: Audiola-linux-x64-AppImage

Audiola and Singola update themselves automatically on launch, powered by Velopack.

---

Self-provisioning & GPU / CUDA

Audiola sets up everything its AI features need — no manual pip install required.

• A managed, isolated Python virtual environment is created under your local app data from the base Python you point it at in Settings.
• Each model is downloaded on demand with one click (Demucs, audio-separator, faster-whisper, the local voice models, seed-vc, etc.), installed into that managed environment.
• In the Voices page you can pick the compute device (auto / cuda / cpu / directml), run a GPU check to see whether CUDA is active and which GPU is detected, and click Install CUDA Torch to install a matching CUDA build of PyTorch for GPU acceleration.

If you'd rather wire up Python yourself, point Audiola's Python path at any interpreter (python, py, or a venv path) in Settings.

---

Build & run from source

Requires the .NET 10 SDK on Windows, macOS, or Linux.

# Clone
git clone https://github.com/fgilde/Audiola.git
cd Audiola

# Run the Avalonia desktop app
dotnet run --project src/Audiola.Avalonia/Audiola.Avalonia.csproj

Build the whole solution:

dotnet build Audiola.sln -c Release

Solution layout

Project	Target	Purpose
src/Audiola.Avalonia/Audiola.Avalonia.csproj	net10.0	Cross-platform Avalonia desktop studio app.
src/Singola.Avalonia/Singola.Avalonia.csproj	net10.0	Cross-platform Avalonia karaoke app.
src/Audiola.Core/Audiola.Core.csproj	net10.0	Shared DSP, mastering, stem separation, voices, project format.
src/Audiola.Web/Audiola.Web.csproj	net10.0	Blazor WebAssembly web companion.
src/Audiola.Api/Audiola.Api.csproj	net10.0	API that hosts the WASM client (same origin).
src/Audiola.AppHost/Audiola.AppHost.csproj	net10.0	.NET Aspire app host for the web stack.

---

Web version

A .NET 10 Blazor WebAssembly companion (src/Audiola.Web) is hosted by src/Audiola.Api, orchestrated for local development by the .NET Aspire app host (src/Audiola.AppHost).

🔗 Open Web version: hosting URL coming soon.

---

Tech stack

.NET 10 · Avalonia (Fluent) · NAudio + NAudio.Lame · SoundTouch · CommunityToolkit.Mvvm · TagLibSharp · Velopack (auto-update) · .NET 10 Blazor WebAssembly + .NET Aspire · MudBlazor.Extensions · AuralizeBlazor.

Python sidecars power the AI/ML parts: Demucs, audio-separator, faster-whisper, seed-vc, Kokoro, Qwen3-TTS, Coqui XTTS v2 and Chatterbox.

---

Releasing

Pushing a v*.*.* tag triggers the GitHub Actions workflow (.github/workflows/release.yml), which publishes the app, builds the Velopack installer and attaches it (plus the update feed) to a GitHub release:

git tag v0.1.0
git push origin v0.1.0

---

Documentation site

The full marketing & documentation site lives in docs/ and is published with GitHub Pages:

➡️ https://fgilde.github.io/Audiola/

---

License

© Florian Gilde. All rights reserved.

This project is currently distributed without an open-source license; please see the repository or contact the author for usage terms.