Building FloAud.io in Public: Why I'm Betting on Pay-Per-Use Audio Processing
After 20 years in post-production and a decade building cloud infrastructure, I'm combining both worlds into something that's never existed before — and sharing the whole messy process.
I've spent two decades in post-production — as a producer, editor, and VFX artist — and the last ten years building cloud infrastructure for studios. At some point it became obvious that the audio software world is stuck in a model that no longer makes sense.
Professional audio processing tools are either:
- Expensive perpetual licenses ($179–$499+ per plugin) that require a local install
- Monthly subscriptions ($22–$299/month) you pay whether you process anything or not
- Consumer tools that don't meet broadcast standards
There's no middle ground for the developer who wants to add professional audio processing to their app, or the podcaster who processes 30 minutes a week and doesn't need a $399 iZotope license sitting idle.
That's the gap FloAud.io fills.
What It Is
FloAud.io is a browser-native, node-based audio processing platform. The core idea: professional-grade audio effects — AI noise reduction, EQ mastering, LUFS normalization, multi-format transcoding — accessible via API or browser interface, priced at $0.10/minute of processed audio.
No downloads. No subscriptions. No minimum commitment.
The browser interface is a node-based editor (think Blender's shader editor, but for audio). You chain effects together visually, save the chain as a preset, and either use it interactively or deploy it via API.
The API looks like this:
curl -X POST https://api.floaud.io/v1/process \
-H "Authorization: Bearer YOUR_API_KEY" \
-F "audio=@podcast-episode.mp3" \
-F "chain=podcast-optimizer" \
-F "output_format=mp3"That's it. Your audio goes in, processed audio comes out, you're billed for the minutes processed.
Why Node-Based
The node-based interface isn't just aesthetically interesting — it maps directly to how audio processing actually works. Every effect is a transform on a signal. Chaining transforms is composition. Branching (processing a signal two different ways and mixing them back) is trivially representable as a graph but awkward in a linear effects chain.
More practically: node graphs are serializable. A chain you build in the browser becomes a JSON document that the API can execute server-side, with identical results. The visual editor is a authoring tool for API payloads.
{
"chain": [
{
"effect": "rnnoise",
"params": { "model": "voice-clarity-v2", "strength": 0.85 }
},
{
"effect": "eq",
"params": { "preset": "podcast-warmth" }
},
{
"effect": "lufs_normalize",
"params": { "target": -16.0, "true_peak": -1.5 }
}
]
}The Technical Stack
The backend is AWS-native because that's where my expertise lives and because the scalability story is straightforward.
Processing workers: EC2 instances running Python with Pedalboard (Spotify's audio processing library) and FFmpeg. Pedalboard gives us VST3 plugin support and professional-grade DSP; FFmpeg handles transcoding and format conversion.
AI noise reduction: We integrate 5 RNNoise models, ranging from aggressive (good for heavily contaminated recordings) to gentle (for subtle cleanup on already-decent audio). The model selection is exposed as a parameter.
Job queue: SQS for async processing. Short jobs (under 30 seconds) run synchronously via API. Longer jobs return a job ID you poll for completion.
Frontend: React + TypeScript with a custom node graph renderer. The graph editor was the hardest part to build — turns out implementing drag-and-drop node connections with correct edge routing is genuinely non-trivial.
Auth and billing: Standard JWT auth, credits system for billing. Users buy credits; processing consumes them at $0.10/minute of audio.
What's Shipping in 2026
The core processing API and browser interface are built. What's left for the July 2026 launch:
- VST3 plugin: Process audio from your DAW using FloAud.io's API. The plugin appears in Logic, Pro Tools, Ableton, etc. as a standard audio effect, but the processing happens in the cloud.
- Browser DAW: A basic multitrack editor with FloAud.io processing built in. Not trying to replace Logic — but enough to record, clean up, and export a podcast episode without leaving the browser.
- Direct DAW API integration: Companion plugin that exposes FloAud.io presets directly in your existing DAW's effect chain.
Why Build in Public
I'm writing about the process because I spent months unable to find any documentation on what I was trying to do. The Pedalboard + RNNoise integration, the node graph rendering approach, the pricing model — I had to figure all of it out through experimentation.
If I document it, maybe someone else doesn't have to.
Also: building in public creates accountability. It's much harder to delay a launch when you've told 1,600 LinkedIn followers it's coming.
The Honest Part
I've been a systems engineer for a decade. I know how to build infrastructure that scales. I do not know how to market a SaaS product, and I'm learning in real time.
The thing I keep coming back to: the pricing model is either the whole insight or it's the whole problem. If developers and creators see $0.10/minute as obviously better than subscriptions, this works. If they don't trust pay-per-use billing, or they don't want to think about cost per use, it won't matter how good the processing is.
I'm betting it works. We'll see.
FloAud.io launches July 2026. If you want early access, DMs are open.