fix(denoise-tester): play models in both ears; match the call's DeepFilterNet strength
CI / Build & Quality Checks (push) Skipped
CI / Secret scan (gitleaks) (push) Skipped
CI / Playwright smoke (e2e) (push) Skipped
CI / Secret scan (gitleaks) (pull_request) Successful in 21s
CI / Build & Quality Checks (pull_request) Successful in 7m3s
CI / Trigger Desktop Build (pull_request) Skipped
CI / Docker image build & smoke test (pull_request) Skipped
CI / Playwright smoke (e2e) (pull_request) Successful in 27m13s
CI / Build & Quality Checks (push) Skipped
CI / Secret scan (gitleaks) (push) Skipped
CI / Playwright smoke (e2e) (push) Skipped
CI / Secret scan (gitleaks) (pull_request) Successful in 21s
CI / Build & Quality Checks (pull_request) Successful in 7m3s
CI / Trigger Desktop Build (pull_request) Skipped
CI / Docker image build & smoke test (pull_request) Skipped
CI / Playwright smoke (e2e) (pull_request) Successful in 27m13s
- Recorded clips often decode as stereo (Chrome Opus/WebM). The models only process channel 0, and the default channel mode ignores channelCount, so RNNoise/Speex played the cleaned audio in the LEFT EAR ONLY. Downmix to mono before the models (as Lotus Call does, #24) and pin the RNNoise/Speex/gate nodes to explicit mono. Verified in Chromium with a stereo speech clip through RNNoise: before L 16.1 / R 0.0, after L 16.0 / R 15.9. - DeepFilterNet ran at attenuation limit 80 here vs 60 in calls; both muted the background outright, which sounds like the mic cutting out at word ends. Use 25, matching element-call#65. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017PCb59uY3UKqDqt7B6hzd5
This commit is contained in:
co-authored by
Claude Opus 5.5
parent
d1a9625b1c
commit
2f5bcf2b91
@@ -6,6 +6,7 @@ import {
|
|||||||
DenoiseNode,
|
DenoiseNode,
|
||||||
buildGateNode,
|
buildGateNode,
|
||||||
buildModelNode,
|
buildModelNode,
|
||||||
|
monoDownmix,
|
||||||
readDb,
|
readDb,
|
||||||
sampleRateFor,
|
sampleRateFor,
|
||||||
} from '../../../utils/denoisePipeline';
|
} from '../../../utils/denoisePipeline';
|
||||||
@@ -139,7 +140,7 @@ export function DenoiseTester({ model, useGate, gateThreshold, nativeNS }: Denoi
|
|||||||
inAnalyser.fftSize = 1024;
|
inAnalyser.fftSize = 1024;
|
||||||
source.connect(inAnalyser);
|
source.connect(inAnalyser);
|
||||||
|
|
||||||
let head: AudioNode = source;
|
let head: AudioNode = source.connect(monoDownmix(ctx));
|
||||||
let gate: AudioWorkletNode | null = null;
|
let gate: AudioWorkletNode | null = null;
|
||||||
if (useGate) {
|
if (useGate) {
|
||||||
gate = await buildGateNode(ctx, gateThreshold);
|
gate = await buildGateNode(ctx, gateThreshold);
|
||||||
@@ -308,10 +309,11 @@ export function DenoiseTester({ model, useGate, gateThreshold, nativeNS }: Denoi
|
|||||||
const ctx = new AudioContext({ sampleRate: sampleRateFor(playModel ?? 'rnnoise') });
|
const ctx = new AudioContext({ sampleRate: sampleRateFor(playModel ?? 'rnnoise') });
|
||||||
const source = ctx.createBufferSource();
|
const source = ctx.createBufferSource();
|
||||||
source.buffer = clip;
|
source.buffer = clip;
|
||||||
|
const mono = source.connect(monoDownmix(ctx));
|
||||||
let playGate: AudioWorkletNode | null = null;
|
let playGate: AudioWorkletNode | null = null;
|
||||||
let playModelNode: DenoiseNode | null = null;
|
let playModelNode: DenoiseNode | null = null;
|
||||||
if (playModel) {
|
if (playModel) {
|
||||||
let head: AudioNode = source;
|
let head: AudioNode = mono;
|
||||||
if (useGate) {
|
if (useGate) {
|
||||||
playGate = await buildGateNode(ctx, gateThreshold);
|
playGate = await buildGateNode(ctx, gateThreshold);
|
||||||
head.connect(playGate);
|
head.connect(playGate);
|
||||||
@@ -321,7 +323,7 @@ export function DenoiseTester({ model, useGate, gateThreshold, nativeNS }: Denoi
|
|||||||
head.connect(playModelNode.node);
|
head.connect(playModelNode.node);
|
||||||
playModelNode.node.connect(ctx.destination);
|
playModelNode.node.connect(ctx.destination);
|
||||||
} else {
|
} else {
|
||||||
source.connect(ctx.destination);
|
mono.connect(ctx.destination);
|
||||||
}
|
}
|
||||||
// Superseded while the WASM/worklet loaded (another Play, a Stop, or the
|
// Superseded while the WASM/worklet loaded (another Play, a Stop, or the
|
||||||
// panel unmounted)? Tear down this now-orphaned graph instead of storing
|
// panel unmounted)? Tear down this now-orphaned graph instead of storing
|
||||||
|
|||||||
@@ -21,6 +21,20 @@ const BASE = `${import.meta.env.BASE_URL.replace(/\/+$/, '')}/public/element-cal
|
|||||||
*/
|
*/
|
||||||
export const sampleRateFor = (model: DenoiseModelId): number => (model === 'dtln' ? 16000 : 48000);
|
export const sampleRateFor = (model: DenoiseModelId): number => (model === 'dtln' ? 16000 : 48000);
|
||||||
|
|
||||||
|
/**
|
||||||
|
* A pass-through node that downmixes whatever feeds it to ONE channel. Every
|
||||||
|
* model only processes channel 0, and a recorded clip often decodes as stereo
|
||||||
|
* (Chrome's Opus/WebM), so without this RNNoise/Speex played the cleaned audio
|
||||||
|
* in the left ear only. Lotus Call does the same (#24). A mono output is then
|
||||||
|
* upmixed to both ears by the destination.
|
||||||
|
*/
|
||||||
|
export const monoDownmix = (ctx: BaseAudioContext): GainNode =>
|
||||||
|
new GainNode(ctx, {
|
||||||
|
channelCount: 1,
|
||||||
|
channelCountMode: 'explicit',
|
||||||
|
channelInterpretation: 'speakers',
|
||||||
|
});
|
||||||
|
|
||||||
export type DenoiseNode = {
|
export type DenoiseNode = {
|
||||||
node: AudioWorkletNode;
|
node: AudioWorkletNode;
|
||||||
dispose: () => void;
|
dispose: () => void;
|
||||||
@@ -85,7 +99,10 @@ export async function buildModelNode(
|
|||||||
const mod: any = await import(/* @vite-ignore */ `${BASE}deepfilternet/index.esm.js`);
|
const mod: any = await import(/* @vite-ignore */ `${BASE}deepfilternet/index.esm.js`);
|
||||||
const core = new mod.DeepFilterNet3Core({
|
const core = new mod.DeepFilterNet3Core({
|
||||||
sampleRate: sampleRateFor(model),
|
sampleRate: sampleRateFor(model),
|
||||||
noiseReductionLevel: 80,
|
// Same attenuation limit as Lotus Call (dB). Higher values mute the
|
||||||
|
// background outright, so words drop into dead silence and sound
|
||||||
|
// clipped at the end.
|
||||||
|
noiseReductionLevel: 25,
|
||||||
assetConfig: { cdnUrl: dfnBase },
|
assetConfig: { cdnUrl: dfnBase },
|
||||||
});
|
});
|
||||||
await core.initialize();
|
await core.initialize();
|
||||||
@@ -105,6 +122,8 @@ export async function buildModelNode(
|
|||||||
const [, wasmBinary] = await Promise.all([addModuleOnce(ctx, cfg.script), fetchWasm(cfg.wasm)]);
|
const [, wasmBinary] = await Promise.all([addModuleOnce(ctx, cfg.script), fetchWasm(cfg.wasm)]);
|
||||||
const node = new AudioWorkletNode(ctx, cfg.proc, {
|
const node = new AudioWorkletNode(ctx, cfg.proc, {
|
||||||
channelCount: 1,
|
channelCount: 1,
|
||||||
|
channelCountMode: 'explicit',
|
||||||
|
channelInterpretation: 'speakers',
|
||||||
numberOfInputs: 1,
|
numberOfInputs: 1,
|
||||||
numberOfOutputs: 1,
|
numberOfOutputs: 1,
|
||||||
processorOptions: { maxChannels: 1, wasmBinary },
|
processorOptions: { maxChannels: 1, wasmBinary },
|
||||||
@@ -127,6 +146,9 @@ export async function buildGateNode(
|
|||||||
): Promise<AudioWorkletNode> {
|
): Promise<AudioWorkletNode> {
|
||||||
await addModuleOnce(ctx, 'noiseGateWorklet.js');
|
await addModuleOnce(ctx, 'noiseGateWorklet.js');
|
||||||
return new AudioWorkletNode(ctx, '@sapphi-red/web-noise-suppressor/noise-gate', {
|
return new AudioWorkletNode(ctx, '@sapphi-red/web-noise-suppressor/noise-gate', {
|
||||||
|
channelCount: 1,
|
||||||
|
channelCountMode: 'explicit',
|
||||||
|
channelInterpretation: 'speakers',
|
||||||
processorOptions: {
|
processorOptions: {
|
||||||
openThreshold: thresholdDb,
|
openThreshold: thresholdDb,
|
||||||
closeThreshold: thresholdDb - 5,
|
closeThreshold: thresholdDb - 5,
|
||||||
|
|||||||
Reference in New Issue
Block a user