Aria Bridge

Aria Bridge connects an AI music model to your DAW in real time. You play a few bars into it, it generates a continuation, and plays it back through a virtual MIDI port — all without leaving your session.

Supported platforms: Windows 10/11 and macOS (Apple Silicon and Intel).

---

What You Need Before Starting

• Ableton Live, Reaper, or any DAW that supports virtual MIDI ports
• An NVIDIA GPU (Windows) or Apple Silicon Mac is recommended for fast generation — CPU and Intel Mac work but are slower
• About 10–15 minutes for a first-time setup

---

Step 1 — Download Aria Bridge

Go to the Releases page and download the latest zip for your platform:

• Windows: AriaBridge-vX.X.X-windows.zip
• macOS: AriaBridge-vX.X.X-macos.zip

Unzip it to a permanent folder — for example C:\Aria Bridge\ on Windows or ~/Aria Bridge/ on Mac.

Do not run it from your Downloads folder. Keep it somewhere stable.

---

Step 2 — Download the Python Backend (Windows only)

The Windows backend (aria_backend.exe) is too large to include in the release zip. You need to download it separately.

1. Go to the Aria Bridge releases on HuggingFace
2. Download aria_backend.exe
3. Place it directly inside your Aria Bridge folder, next to the plugin

Aria Bridge/
  Aria Bridge.exe    (standalone plugin)
  Aria Bridge.vst3   (VST3 plugin)
  aria_backend.exe   ← goes here
  models/
  ableton/
  ...

macOS users can skip this step — the backend is already included in the Mac zip.

---

Step 3 — Download the AI Model

The model file is not included because of its size. You need to download it once.

1. Download: model-gen.safetensors
2. Place it inside the models folder in your Aria Bridge folder

Aria Bridge/
  models/
    model-gen.safetensors   ← goes here

---

Step 4 — Set Up Virtual MIDI Ports

Aria Bridge talks to your DAW through virtual MIDI cables.

Windows — loopMIDI

1. Download loopMIDI (free): tobias-erichsen.de/software/loopmidi.html
2. Install and open it
3. Create two ports with these exact names:
   • ARIA_IN
   • ARIA_OUT

loopMIDI must be running every time you use Aria Bridge. You can set it to launch on startup in its settings.

macOS — IAC Driver (built in)

macOS has a virtual MIDI system built in — no extra download needed.

1. Open Audio MIDI Setup (search in Spotlight)
2. Go to Window → Show MIDI Studio
3. Double-click IAC Driver
4. Check Device is online
5. Add two ports named exactly:
   • ARIA_IN
   • ARIA_OUT

---

Step 5 — Route MIDI in Your DAW

You need two MIDI tracks:

Track	Purpose	MIDI Output / Input
Track 1	Your instrument (what you play)	Output → ARIA_IN
Track 2	Aria's output (what it generates)	Input → ARIA_OUT

Put an instrument on Track 2 so you can hear Aria's output.

---

Step 6 — Launch Aria Bridge

Note: the one-click graphical launcher is being rebuilt. For now you start the backend from a terminal, then use the plugin or Max for Live device. Everything below works the same once it's running.

1. Start the backend

Open a terminal in your Aria Bridge folder and run:

• Windows: .\aria_backend.exe plugin --checkpoint models\model-gen.safetensors
• macOS: ./aria_backend plugin --checkpoint models/model-gen.safetensors

The backend auto-detects your GPU (CUDA on Windows, MLX on Apple Silicon, else CPU). When it prints STATUS:ready, it's connected. Leave this window open while you play.

2. Use it from your DAW

• M4L Device — load the included Max for Live device inside Ableton, or
• Plugin (VST3 / Standalone) — load Aria Bridge.vst3 in any DAW, or open the Aria Bridge standalone.

The plugin/device talks to the running backend over OSC.

---

Using Aria Bridge

Control	What it does
record	Start / stop recording your MIDI input
play	Play back the generated output
cancel	Stop whatever is happening — cancels recording, interrupts generation, stops playback, or discards a pending output and returns to record (also stops the loop)
commit	Save this generation with your ratings
sync	Re-send all parameter values to the backend
temp	Temperature — higher = more surprising, lower = more conservative
top_p / min_p	Sampling filters — leave at defaults to start
tokens	How many tokens to generate — more = longer output
clip	Send the generation into an Ableton clip (via AbletonOSC) instead of out ARIA_OUT
loop	Infinite loop — feed each output back as the next prompt, clip after clip (needs clip on)
track / slot	Which Ableton track / clip slot to write into
fire	Auto-launch each clip as it's written (off = you launch them)
coherence / taste / repetition / continuity / grade	Rate the generation 1–5 before committing

Basic workflow:

1. Hit record and play something on your MIDI track
2. Hit record again to stop — generation starts automatically
3. A timer shows while the model is running
4. When ready, hit play to hear the result
5. If you like it, rate it and hit commit

---

Ableton clip output & infinite loop (advanced)

Instead of playing the generation out a MIDI port, Aria can drop it straight into an Ableton clip so Live's own engine plays it — perfectly in time and re-triggerable. This needs AbletonOSC, a free control surface for Ableton.

Install AbletonOSC (one time)

1. Download it from GitHub: github.com/ideoforms/AbletonOSC → green Code button → Download ZIP, then unzip. (You want the inner AbletonOSC folder.)
2. Move the AbletonOSC folder into Ableton's Remote Scripts folder (create the folder if it doesn't exist):
   • Windows: \Users\<you>\Documents\Ableton\User Library\Remote Scripts\
   • macOS: ~/Music/Ableton/User Library/Remote Scripts/
3. Restart Ableton Live.
4. Preferences → Link/Tempo/MIDI, under Control Surface pick AbletonOSC, and set Input = None and Output = None (it uses OSC over the network, not MIDI).

That's it — Live's log will show AbletonOSC: Listening for OSC on port 11000 when it loads.

Use it

• clip on → each generation is written into the track/slot set by track / slot.
• loop on (needs clip) → infinite mode: record one idea to seed it, then Aria keeps feeding its own output back in and writing clip after clip down the column. You launch the clips in Ableton; cancel stops the loop.
• fire on → clips auto-launch as they're written (off by default in loop, so you trigger them).

Tip: set Ableton's Global Launch Quantization to 1 Bar so clips start and chain on the downbeat.

---

Troubleshooting

Status shows DISCONNECTED The backend isn't running. Start it from a terminal as in Step 6 and leave that window open. On Windows, make sure aria_backend.exe was downloaded from HuggingFace and placed in your Aria Bridge folder (see Step 2). Check the terminal for an error if it exits immediately.

No MIDI is being captured Check that your virtual MIDI ports exist and are named exactly ARIA_IN and ARIA_OUT. On Windows, make sure loopMIDI is running. Check your DAW track routing.

Generation is very slow On Windows, make sure you have an NVIDIA GPU with up-to-date drivers. On Mac, this is expected on Intel — Apple Silicon is significantly faster.

The model file is not found Make sure model-gen.safetensors is inside the models folder and has not been renamed.

---

License

MIT