feat(settings): make DTLN the default ML noise-suppression model #482
@@ -2037,7 +2037,7 @@ export function Calls() {
|
|||||||
<Text size="L400">Model</Text>
|
<Text size="L400">Model</Text>
|
||||||
<SettingTile
|
<SettingTile
|
||||||
title="ML model"
|
title="ML model"
|
||||||
description="Choose the machine learning model used for noise removal. Heavier models clean more aggressively at a higher CPU cost. CPU figures are the share of one processor core during a call on a mid-range desktop; Task Manager shows that divided by your number of cores, and slower computers use more."
|
description="Choose the machine learning model used for noise removal. DTLN is the best all-rounder; DeepFilterNet removes the most noise at a higher CPU cost. None of the models remove other people talking in the background. CPU figures are the share of one processor core during a call on a mid-range desktop; Task Manager shows that divided by your number of cores, and slower computers use more."
|
||||||
after={
|
after={
|
||||||
<SettingsSelect<DenoiseModelId>
|
<SettingsSelect<DenoiseModelId>
|
||||||
value={callDenoiseModel}
|
value={callDenoiseModel}
|
||||||
@@ -2066,7 +2066,7 @@ export function Calls() {
|
|||||||
[
|
[
|
||||||
{ label: 'CPU (one core)', value: selectedDenoiseModel.cpuUsage },
|
{ label: 'CPU (one core)', value: selectedDenoiseModel.cpuUsage },
|
||||||
{ label: 'Voice quality', value: selectedDenoiseModel.voiceQuality },
|
{ label: 'Voice quality', value: selectedDenoiseModel.voiceQuality },
|
||||||
{ label: 'Transients', value: selectedDenoiseModel.transients },
|
{ label: 'Typing & clicks', value: selectedDenoiseModel.transients },
|
||||||
{ label: 'Download', value: selectedDenoiseModel.binarySize },
|
{ label: 'Download', value: selectedDenoiseModel.binarySize },
|
||||||
] as const
|
] as const
|
||||||
).map((stat) => (
|
).map((stat) => (
|
||||||
@@ -2113,7 +2113,7 @@ export function Calls() {
|
|||||||
</Box>
|
</Box>
|
||||||
<Box grow="Yes">
|
<Box grow="Yes">
|
||||||
<Text size="T200" priority="300">
|
<Text size="T200" priority="300">
|
||||||
Transients
|
Typing & clicks
|
||||||
</Text>
|
</Text>
|
||||||
</Box>
|
</Box>
|
||||||
</Box>
|
</Box>
|
||||||
|
|||||||
@@ -377,9 +377,11 @@ const defaultSettings: Settings = {
|
|||||||
|
|
||||||
cameraOnJoin: false,
|
cameraOnJoin: false,
|
||||||
// Tier default stays browser-native (known-good; best-perceived in testing so
|
// Tier default stays browser-native (known-good; best-perceived in testing so
|
||||||
// far). If a user opts into the ML tier, default to the highest-quality model.
|
// far). If a user opts into the ML tier, default to DTLN: in our measurements
|
||||||
|
// it keeps voices the most natural, removes fans and typing well, and costs a
|
||||||
|
// third of DeepFilterNet's CPU.
|
||||||
callNoiseSuppression: 'browser',
|
callNoiseSuppression: 'browser',
|
||||||
callDenoiseModel: 'deepfilternet',
|
callDenoiseModel: 'dtln',
|
||||||
// "Series suppression" (stack the browser's native NS before the ML model) is
|
// "Series suppression" (stack the browser's native NS before the ML model) is
|
||||||
// off by default — best practice is a single NS stage; it's an opt-in test aid.
|
// off by default — best practice is a single NS stage; it's an opt-in test aid.
|
||||||
callDenoiseNativeNS: false,
|
callDenoiseNativeNS: false,
|
||||||
|
|||||||
@@ -5,10 +5,10 @@ import { DENOISE_MODELS, ML_DENOISE_REQUIREMENTS, isMLDenoiseSupported } from '.
|
|||||||
|
|
||||||
// ── Model catalog (data integrity) ──────────────────────────────────────────
|
// ── Model catalog (data integrity) ──────────────────────────────────────────
|
||||||
|
|
||||||
test('DENOISE_MODELS lists the four models ordered best-quality (highest CPU) first', () => {
|
test('DENOISE_MODELS lists the four models most-recommended first', () => {
|
||||||
assert.deepEqual(
|
assert.deepEqual(
|
||||||
DENOISE_MODELS.map((m) => m.id),
|
DENOISE_MODELS.map((m) => m.id),
|
||||||
['deepfilternet', 'dtln', 'rnnoise', 'speex'],
|
['dtln', 'deepfilternet', 'rnnoise', 'speex'],
|
||||||
);
|
);
|
||||||
});
|
});
|
||||||
|
|
||||||
|
|||||||
@@ -1,8 +1,8 @@
|
|||||||
/**
|
/**
|
||||||
* Detection utilities + model catalog for Lotus ML noise suppression
|
* Detection utilities + model catalog for Lotus ML noise suppression
|
||||||
* (DeepFilterNet 3 / DTLN / RNNoise / Speex). The catalog is ordered by
|
* (DTLN / DeepFilterNet 3 / RNNoise / Speex). The catalog is ordered by
|
||||||
* quality (and, correspondingly, CPU cost) — highest first — and drives the
|
* recommendation — most recommended first — and drives the order of the model
|
||||||
* order of the model dropdown in settings.
|
* dropdown in settings.
|
||||||
*/
|
*/
|
||||||
|
|
||||||
import { DenoiseModelId } from '../state/settings';
|
import { DenoiseModelId } from '../state/settings';
|
||||||
@@ -10,6 +10,9 @@ import { DenoiseModelId } from '../state/settings';
|
|||||||
// CPU = share of ONE core while in a call, measured with each model running
|
// CPU = share of ONE core while in a call, measured with each model running
|
||||||
// exactly as Lotus Call runs it (Ryzen 7 PRO 4750GE, Chromium); download = bytes
|
// exactly as Lotus Call runs it (Ryzen 7 PRO 4750GE, Chromium); download = bytes
|
||||||
// actually fetched on first use. Re-measure when a model or its settings change.
|
// actually fetched on first use. Re-measure when a model or its settings change.
|
||||||
|
// Ratings come from scoring each model against the clean voice (PESQ/STOI) on
|
||||||
|
// real speech mixed with fan, keyboard, street and babble noise at SNR 5 and
|
||||||
|
// 15 dB; `transients` is how well it removes typing and clicks.
|
||||||
export type DenoiseModel = {
|
export type DenoiseModel = {
|
||||||
id: DenoiseModelId;
|
id: DenoiseModelId;
|
||||||
name: string;
|
name: string;
|
||||||
@@ -20,25 +23,25 @@ export type DenoiseModel = {
|
|||||||
voiceQuality: 'Moderate' | 'High' | 'Very High';
|
voiceQuality: 'Moderate' | 'High' | 'Very High';
|
||||||
};
|
};
|
||||||
|
|
||||||
// Ordered best-quality (highest CPU) first — this is the dropdown order.
|
// Ordered most-recommended first — this is the dropdown order.
|
||||||
export const DENOISE_MODELS: DenoiseModel[] = [
|
export const DENOISE_MODELS: DenoiseModel[] = [
|
||||||
{
|
{
|
||||||
id: 'deepfilternet',
|
id: 'dtln',
|
||||||
name: 'DeepFilterNet 3 (beta)',
|
name: 'DTLN (recommended)',
|
||||||
description:
|
description:
|
||||||
'Studio-grade deep-learning model (48 kHz fullband, ONNX). Best quality; highest CPU and a larger one-time download.',
|
'Keeps voices the most natural of the four. Removes fans and typing well, but not traffic or people talking in the background.',
|
||||||
cpuUsage: '~30%',
|
cpuUsage: '~9%',
|
||||||
binarySize: '~18 MB',
|
binarySize: '~20 MB',
|
||||||
transients: 'Excellent',
|
transients: 'Excellent',
|
||||||
voiceQuality: 'Very High',
|
voiceQuality: 'Very High',
|
||||||
},
|
},
|
||||||
{
|
{
|
||||||
id: 'dtln',
|
id: 'deepfilternet',
|
||||||
name: 'DTLN (beta)',
|
name: 'DeepFilterNet 3 (strongest)',
|
||||||
description:
|
description:
|
||||||
'Dual-signal deep-learning model (16 kHz). Strong on transient noise; moderate CPU.',
|
'Removes the most noise, including loud typing and traffic, but voices can sound slightly muffled. Uses about three times the CPU of DTLN.',
|
||||||
cpuUsage: '~9%',
|
cpuUsage: '~30%',
|
||||||
binarySize: '~20 MB',
|
binarySize: '~18 MB',
|
||||||
transients: 'Excellent',
|
transients: 'Excellent',
|
||||||
voiceQuality: 'High',
|
voiceQuality: 'High',
|
||||||
},
|
},
|
||||||
@@ -46,17 +49,17 @@ export const DENOISE_MODELS: DenoiseModel[] = [
|
|||||||
id: 'rnnoise',
|
id: 'rnnoise',
|
||||||
name: 'RNNoise',
|
name: 'RNNoise',
|
||||||
description:
|
description:
|
||||||
'Lightweight hybrid model (48 kHz). Very low CPU; good for steady noise like fans, but can sound processed at full strength.',
|
'Very light. Handles steady noise like fans and traffic; does little for typing and can sound slightly processed.',
|
||||||
cpuUsage: '~3%',
|
cpuUsage: '~3%',
|
||||||
binarySize: '< 1 MB',
|
binarySize: '< 1 MB',
|
||||||
transients: 'Good',
|
transients: 'Poor',
|
||||||
voiceQuality: 'Moderate',
|
voiceQuality: 'Moderate',
|
||||||
},
|
},
|
||||||
{
|
{
|
||||||
id: 'speex',
|
id: 'speex',
|
||||||
name: 'Speex (Legacy)',
|
name: 'Speex (legacy)',
|
||||||
description:
|
description:
|
||||||
'Classic DSP noise suppressor. Minimal CPU, gentlest on voice; weakest suppression.',
|
'Classic suppressor kept for very slow computers. Barely reduces noise in our tests; prefer any other option.',
|
||||||
cpuUsage: '~2%',
|
cpuUsage: '~2%',
|
||||||
binarySize: '< 1 MB',
|
binarySize: '< 1 MB',
|
||||||
transients: 'Poor',
|
transients: 'Poor',
|
||||||
|
|||||||
Reference in New Issue
Block a user