@nanoya-biz/tauri-plugin-noise-suppression (0.1.4)
Installation
@nanoya-biz:registry=npm config set @nanoya-biz:registry=npm install @nanoya-biz/tauri-plugin-noise-suppression@0.1.4"@nanoya-biz/tauri-plugin-noise-suppression": "0.1.4"About this package
tauri-plugin-noise-suppression
Tauri 2.x plugin that exposes rkd-noise-suppression-core::Denoiser to the
webview via Tauri commands.
Status — Phase 3b. Plugin built and type-checked; demo scaffold committed. Awaiting manual run-through validation (
npm run tauri:devinexamples/tauri-demo/). Do not depend on this in production until that validation step has been completed and documented.
Build
The plugin has two parts: a Rust crate and a JS wrapper. The Rust crate is
built automatically by Cargo when your Tauri app compiles. The JS wrapper must
be built once manually so the dist-js/ directory exists for the demo or any
consuming app.
cd crates/tauri-plugin
npm install
npm run build
This produces dist-js/{index.js, index.mjs, index.d.ts} from the TypeScript
source in guest-js/index.ts.
Add to a Tauri 2.x app
1. Cargo dependency
In src-tauri/Cargo.toml:
[dependencies]
tauri-plugin-noise-suppression = { path = "<relative-path-to>/crates/tauri-plugin" }
Once the plugin is published to crates.io (Phase 3b-2), this becomes:
tauri-plugin-noise-suppression = "0.1"
2. JS dependency
In your app's package.json:
{
"dependencies": {
"tauri-plugin-noise-suppression": "file:../../crates/tauri-plugin"
}
}
Build the JS wrapper first (npm install && npm run build in
crates/tauri-plugin/) so the dist-js/ output exists before your app's
npm install resolves the file dep.
3. Register the plugin
In src-tauri/src/main.rs:
fn main() {
tauri::Builder::default()
.plugin(tauri_plugin_noise_suppression::init())
// … other plugins …
.run(tauri::generate_context!())
.expect("error while running tauri application");
}
4. Enable the capability
In src-tauri/capabilities/default.json:
{
"identifier": "default",
"windows": ["main"],
"permissions": [
"core:default",
"noise-suppression:default"
]
}
The noise-suppression:default permission grants all six commands:
init_denoiser, frame_size_at_48k, process_frame, reset,
set_setting, get_setting.
5. Use from frontend code
import {
initDenoiser,
frameSizeAt48k,
processFrame,
reset,
} from "tauri-plugin-noise-suppression";
// Initialize once (before audio capture starts).
await initDenoiser({ modelDir: "/path/to/models/dfn2", sampleRate: 48000 });
// Per-frame call (~100/sec at 48 kHz with 480-sample frames).
const denoised: Float32Array = await processFrame(inputFrame);
// Clean up (e.g. on session end).
await reset();
Setting ORT_DYLIB_PATH
The plugin uses ONNX Runtime via deepfilter-rt. On Windows, the system may
have a stale onnxruntime.dll in System32 that will be loaded in preference
to the version this plugin expects (=2.0.0-rc.11 ORT ABI). Set
ORT_DYLIB_PATH to the bundled runtime before tauri dev starts:
# PowerShell (Windows)
$env:ORT_DYLIB_PATH = "C:\path\to\onnxruntime\lib\onnxruntime.dll"
npm run tauri:dev
# bash / macOS / Linux
ORT_DYLIB_PATH="/path/to/onnxruntime/lib/libonnxruntime.dylib" npm run tauri:dev
Or set it programmatically in src-tauri/src/main.rs before the builder
runs:
fn main() {
// Set before any ORT initialization happens.
std::env::set_var(
"ORT_DYLIB_PATH",
"/absolute/path/to/onnxruntime/lib/onnxruntime.dll",
);
tauri::Builder::default()
.plugin(tauri_plugin_noise_suppression::init())
.run(tauri::generate_context!())
.expect("error while running tauri application");
}
ORT version pin
This plugin pins ort = "=2.0.0-rc.11" — the same version as crates/napi.
Both depend on deepfilter-rt which bundles its own ORT C ABI expectations.
Do not upgrade ORT independently; upgrade both together when deepfilter-rt
releases a new version.
Caveats
-
Per-frame invoke overhead. Each
processFramecall crosses the webview→Tauri IPC boundary, which adds ~1–5 ms on Windows. At 100 frames/sec this is usually inaudible, but on very loaded systems it can cause occasional audio glitches. The SharedArrayBuffer ring-buffer path (Phase 3c) eliminates this overhead. -
Binary IPC. The plugin uses
tauri::ipc::Response::new(bytes)and raw binary request bodies to avoid per-frame JSON serialization. If you invokeprocess_framevia the standard JSON path you will get a "expected raw binary body" error. Always pass aUint8Array/ArrayBufferas the invoke payload. -
Single instance. The current implementation manages one
Denoiserinstance per plugin registration. Multi-mic scenarios (multiple streams denoised simultaneously) require multiple plugin registrations or a future session-keyed API. -
State is app-lifetime.
initDenoiserfails withAlreadyInitializedif called twice. Callresetto clear internal ring buffers without re-initializing, or useinitDenoiserafter app restart.
Reference
See docs/integrations/tauri.md for the
full integration guide including audio graph wiring, bundling, performance
analysis, and troubleshooting.
Dependencies
Development dependencies
| ID | Version |
|---|---|
| @rollup/plugin-typescript | ^12.0.0 |
| @tauri-apps/api | ^2.0.0 |
| rollup | ^4.0.0 |
| tslib | ^2.6.0 |
| typescript | ^5.6.0 |
Peer dependencies
| ID | Version |
|---|---|
| @tauri-apps/api | ^2.0.0 |