Introducing the NAM Web Player
You can now audition any tone on TONE3000 directly in your browser, no downloads required.
Today, we're launching the NAM Web Player site-wide, bringing real-time audio modeling directly to your web browser.
Built for both developers and players, the open-source NAM Web Player lets anyone experience neural modeling on the web. Developers can embed real DSP in their web-apps, and TONE3000 users can instantly browse, compare, and audition tones online before downloading.
What you can do with it:
- Try Before You Download - Hear exactly what a tone sounds like before committing
- Compare Tones Instantly - Audition different tones to find the perfect match for your project
- Test with Your Own Sound - Choose from various guitar and bass DI tracks to hear how each tone responds
- Mix and Match - For amp-only captures, pair them with different speaker cabinet IR's to dial in your sound
- Preview IR's - Audition impulse responses the same way you would any other tone
Every tone on TONE3000 now has a play button. Click it and you're immediately hearing that exact amp, pedal, outboard, or cabinet capture.
Try it out
Here’s a live example of the NAM Web Player running a Vox AC15C1 Top Boost amp head capture by creator @outmodedelectronics:
You’re hearing a real NAM model running natively in your browser — the same model you can download to use in your DAW or pedal.
Customize playback by selecting different DI inputs and, for amp-head-only models, choose from various cabinet impulse responses (IR's).
The player is also great for previewing IR's from the community. Here’s a live example running Marshall 1960AV IR's by creator @pippriss:
Continue auditioning tones live ➝ Browse Tones
How it works
Here's what makes this special: you're not hearing a simple audio preview. The NAM Web Player runs the same neural-network inference engine used in the desktop Neural Amp Modeler: a stack of dilated 1-D convolutional layers (inherited from WaveNet-style modelling) compiled into WebAssembly using Emscripten and executed inside an AudioWorklet via the Web Audio API.
Here’s what’s happening step-by-step:
- Model Loading and Inference
When you load a.namfile, it brings in the networks’s trained weights, layer configuration and scaling data. At playback time, it processes small blocks of input audio through the stack to predict each output sample. This lets the system model the complex, nonlinear response of an amplifier, pedal or rig with very fine temporal resolution. - WebAssembly Runtime
The C++ DSP engine from NeuralAmpModelerCore is compiled into WebAssembly so that the same neural-model-based inference code can run inside a web browser. This gives near-native performance of the neural DSP engine in the browser context. - AudioWorklet Integration
The WASM module runs inside an AudioWorkletProcessor, which communicates with the browser’s audio rendering thread. Audio buffers are streamed in, processed through the NAM model, and streamed back out — all within the same render quantum used by the browser’s audio engine.
No server processing, no backend — just neural DSP, local to your CPU. - Web Player Wrapper
Our open-source frontend handles file loading, model initialization, and browser playback integration, making it easy to embed NAM playback into any site.
Limitations
Because the NAM Web Player is fully local, there are a few technical caveats:
- CORS Limitations
Since the Web Player uses SharedArrayBuffer (via its WASM + AudioWorklet stack), the page must be in a cross-origin-isolated state—i.e., the HTTP response headers must includeCross-Origin-Opener-Policy: same-originandCross-Origin-Embedder-Policy: require-corp. Without this isolation, many AudioWorklet-based real-time DSP paths will fail. - Performance Requirements
Real-time neural inference is a bit CPU-intensive. The player requires a modern processor (Apple Silicon, recent Intel/AMD). Playback may not perform correctly on older or low-power devices. To address this limitation we’re developing a “pre-rendered” version of the player as a fallback to real-time processing.
More utility coming soon
We’re expanding the player into a complete tone utility layer for the web. Upcoming features include:
- Upload your own DI or IR to your account for previewing in the browser
- Input/output gain staging and normalization
- IR mixing for reverb, delay and cabinet blending
- Offline rendering
- A no-code embeddable player for easy integration on parter sites
And much more to come! We welcome your ideas—please submit feature requests and issues on GitHub.
Open source
The NAM Web Player is part of TONE3000’s open-source mission: bringing training and playback into one seamless, interoperable, and accessible ecosystem.
We encourage contributions of:
- Code — improvements to the player
- DI recordings — guitar, bass, drums, vocals, synths, and more for training new models: Input contribution guidelines
- IR's — speaker cabinets and reverbs: IR contribution guidelines
- NAM's — amps, pedals and outboard gear: Model contribution guidelines
Special thanks to:
- Steven Atkinson — for authoring the original NeuralAmpModelerCore
- Kutalia — for earlier work on browser-based NAM playback
- Creator @pippriss for kicking off contributions with three new thrash metal inputs and a EVH 5153 amp-head model.
