Skip to content

armgit / cindy

Public Imported from GitHub
0

Consider it done. The open-source AI agent that works out of the box · 想到,就能做到。开源、开箱即用的 AI Agent。

feat(voice): 默认开启输入框长按语音并提前准备麦克风

Dash98ce549cfb50
diff --git a/apps/desktop/scripts/fixtures/companion-composer/ui-page.mjs b/apps/desktop/scripts/fixtures/companion-composer/ui-page.mjs
index eb02ff1997..18a7bcacc4 100644
--- a/apps/desktop/scripts/fixtures/companion-composer/ui-page.mjs
+++ b/apps/desktop/scripts/fixtures/companion-composer/ui-page.mjs
@@ -24,9 +24,13 @@ export default {
'src/renderer/components/new-chat/NewGoalDialog': `export const NewGoalDialog = () => null;`,
'src/renderer/components/new-chat/AtMentionPanel': `export const AtMentionPanel = () => null;`,
'src/renderer/components/new-chat/SlashCommandPalette': `export const SlashCommandPalette = () => null;`,
- 'src/renderer/voice-input/useVoiceInput': `import { useState } from 'react'; export function useVoiceInput(editor) { const [state, setState] = useState('idle'); return { state, isListening: state === 'listening', isBusy: state !== 'idle', draftText: '', start: () => { setState('listening'); return true; }, stop: async () => { setState('idle'); editor?.commands.setContent('<p>Dictated message</p>'); }, cancel: async () => setState('idle') }; }`,
+ 'src/renderer/voice-input/useVoiceInput': `import { useState, useRef } from 'react'; export function useVoiceInput(editor) { const [state, setState] = useState('idle'); const current = useRef(state); const update = value => { current.current = value; setState(value); }; return { state, getState: () => current.current, isListening: state === 'listening', isBusy: state !== 'idle', draftText: '', start: () => { update('listening'); return true; }, stop: async () => { update('idle'); editor?.commands.setContent('<p>Dictated message</p>'); }, cancel: async () => update('idle') }; }`,
},
cases: {
+ 'voice-hold': {
+ about: 'Empty composer hints, native mouse selection during hold, and button hold/click release semantics',
+ viewport: { width: 360, height: 400 },
+ },
default: {
about:
'Empty/text/attachment action switch, stable target, keyboard send, call return and stop/queue',
@@ -40,6 +44,97 @@ export default {
await page.locator('[contenteditable="true"]').waitFor();
},
async check(page, ui) {
+ if (ui.case === 'voice-hold') {
+ const editor = page.locator('.ProseMirror');
+ const hint = page.getByTestId('composer-voice-hint');
+ const hintText = async (key) => {
+ await hint.waitFor();
+ assert.equal(await hint.textContent(), await ui.t(`newChat.chatInput.voiceInput.${key}`));
+ };
+ const card = page.locator('[data-split-group-composer-drop-target]');
+ const cardBox = await card.boundingBox();
+ // Start in the card's padding, outside the single text line.
+ await page.mouse.move(cardBox.x + 4, cardBox.y + 4);
+ await hintText('holdComposer');
+ await ui.capture('hover');
+ await page.mouse.down();
+ await page.waitForFunction(() => document.querySelector('[data-voice-holding="true"]'));
+ await hintText('speakRelease');
+ await ui.capture('hold-empty');
+ await page.mouse.up();
+ await page.waitForFunction(() => document.querySelector('.ProseMirror')?.textContent === 'Dictated message');
+ assert.equal(await hint.count(), 0);
+
+ // The gap above the toolbar is also part of the input surface.
+ const editorBox = await editor.boundingBox();
+ await page.mouse.move(cardBox.x + cardBox.width / 2, editorBox.y + editorBox.height + 5);
+ await page.mouse.down();
+ await page.waitForFunction(() => document.querySelector('[data-voice-holding="true"]'));
+ await page.mouse.up();
+ await page.locator('.ProseMirror[contenteditable="true"]').waitFor();
+
+ const rect = await editor.boundingBox();
+ const middle = await editor.evaluate((node) => {
+ const text = node.querySelector('p').firstChild;
+ const range = document.createRange();
+ range.setStart(text, 8);
+ range.setEnd(text, 9);
+ const bounds = range.getBoundingClientRect();
+ return { x: bounds.x + bounds.width / 2, y: bounds.y + bounds.height / 2 };
+ });
+ await page.mouse.move(middle.x, middle.y);
+ await page.mouse.down();
+ await page.waitForFunction(() => document.querySelector('[data-voice-holding="true"]'));
+ await page.mouse.move(rect.x + rect.width - 8, rect.y + rect.height / 2, { steps: 12 });
+ assert.equal(await page.evaluate(() => window.getSelection()?.toString()), '');
+ await page.mouse.move(rect.x + 2, middle.y, { steps: 12 });
+ assert.equal(await page.evaluate(() => window.getSelection()?.toString()), '');
+ // Exercise selection changes that escape move/selectstart cancellation
+ // (an already-running native selection gesture). The editor must retain
+ // the insertion point, not adopt the expanded DOM selection.
+ const insertionPoint = await editor.evaluate((node) => node.editor.state.selection.from);
+ await editor.evaluate((node) => {
+ const text = node.querySelector('p').firstChild;
+ document.getSelection().setBaseAndExtent(text, 2, text, 5);
+ });
+ await page.waitForFunction(() => document.getSelection()?.isCollapsed);
+ assert.equal(await editor.evaluate((node) => node.editor.state.selection.from), insertionPoint);
+ assert.equal(await hint.count(), 0);
+ await page.mouse.up();
+ await page.locator('.ProseMirror[contenteditable="true"]').waitFor();
+ assert.equal(await page.evaluate(() => window.getSelection()?.toString()), '');
+ // A queued native selection change can arrive after pointerup/mouseup.
+ // It must still be rejected until a new user interaction begins.
+ await editor.evaluate((node) => {
+ const text = node.querySelector('p').firstChild;
+ document.getSelection().setBaseAndExtent(text, 2, text, 5);
+ });
+ await page.waitForFunction(() => document.getSelection()?.isCollapsed);
+ assert.equal(await editor.evaluate((node) => node.editor.state.selection.empty), true);
+ // Releasing restores ordinary text selection.
+ const releasedText = await editor.locator('p').boundingBox();
+ await page.mouse.move(releasedText.x + 5, releasedText.y + releasedText.height / 2);
+ await page.mouse.down();
+ await page.mouse.move(releasedText.x + 90, releasedText.y + releasedText.height / 2, { steps: 4 });
+ await page.mouse.up();
+ assert.notEqual(await page.evaluate(() => window.getSelection()?.toString()), '');
+ await editor.fill('');
+ const mic = page.locator('button[aria-pressed]');
+ await mic.click();
+ await hintText('speakPressAgain');
+ await ui.capture('click-empty');
+ await mic.click();
+ await page.locator('.ProseMirror[contenteditable="true"]').waitFor();
+ await editor.fill('');
+ await mic.hover();
+ await page.mouse.down();
+ await page.getByRole('button', { name: await ui.t('newChat.chatInput.voiceInput.releaseToStop'), exact: true }).waitFor();
+ await hintText('speakRelease');
+ await ui.capture('button-hold-empty');
+ await page.mouse.up();
+ await page.locator('.ProseMirror[contenteditable="true"]').waitFor();
+ return;
+ }
// Screenshot harness defaults to reduced motion; opt in to verify real transitions.
await page.emulateMedia({ reducedMotion: 'no-preference' });
const editor = page.locator('[contenteditable]');
diff --git a/apps/desktop/src/main/voice-input/__tests__/composerLongPressSetting.test.ts b/apps/desktop/src/main/voice-input/__tests__/composerLongPressSetting.test.ts
index a018665d6b..fd218dfa5e 100644
--- a/apps/desktop/src/main/voice-input/__tests__/composerLongPressSetting.test.ts
+++ b/apps/desktop/src/main/voice-input/__tests__/composerLongPressSetting.test.ts
@@ -65,9 +65,9 @@ afterEach(() => {
});
describe('长按输入框语音输入开关 —— 只持久化用户 override', () => {
- it('默认关闭;无关设置保存不会把默认值记成用户选择', () => {
+ it('默认开启;无关设置保存不会把默认值记成用户选择', () => {
writeSettingsFile({ dictionaryEntries: [] });
- expect(voiceInputDataStore.getSettings().composerLongPressEnabled).toBe(false);
+ expect(voiceInputDataStore.getSettings().composerLongPressEnabled).toBe(true);
voiceInputDataStore.updateSettings({ language: 'zh-CN' });
expect(readPersistedSettings().composerLongPressEnabledOverride).toBeUndefined();
@@ -91,20 +91,22 @@ describe('长按输入框语音输入开关 —— 只持久化用户 override',
voiceInputDataStore.updateSettings({ composerLongPressEnabled: false });
expect(readPersistedSettings().composerLongPressEnabledOverride).toBe(false);
expect(readPersistedSettings().composerLongPressEnabled).toBeUndefined();
+ resetStoreCaches();
+ expect(voiceInputDataStore.getSettings().composerLongPressEnabled).toBe(false);
});
it('传 null 恢复默认:删除 override,重新跟随版本默认值', () => {
writeSettingsFile({ dictionaryEntries: [] });
voiceInputDataStore.updateSettings({ composerLongPressEnabled: true });
const next = voiceInputDataStore.updateSettings({ composerLongPressEnabled: null });
- expect(next.composerLongPressEnabled).toBe(false);
+ expect(next.composerLongPressEnabled).toBe(true);
expect(readPersistedSettings().composerLongPressEnabledOverride).toBeUndefined();
expect(readPersistedSettings().composerLongPressEnabled).toBeUndefined();
});
it('配置里残留的有效值不会被当成用户选择', () => {
// 有效值只是写盘时的投影;没有 override 就一律跟随默认值。
- writeSettingsFile({ dictionaryEntries: [], composerLongPressEnabled: true });
- expect(voiceInputDataStore.getSettings().composerLongPressEnabled).toBe(false);
+ writeSettingsFile({ dictionaryEntries: [], composerLongPressEnabled: false });
+ expect(voiceInputDataStore.getSettings().composerLongPressEnabled).toBe(true);
});
});
diff --git a/apps/desktop/src/renderer/components/new-chat/ChatInput.tsx b/apps/desktop/src/renderer/components/new-chat/ChatInput.tsx
index 976b4a398b..088bb0398f 100644
--- a/apps/desktop/src/renderer/components/new-chat/ChatInput.tsx
+++ b/apps/desktop/src/renderer/components/new-chat/ChatInput.tsx
@@ -3460,6 +3460,7 @@ export function ChatInput({
const voiceShortcutRef = useRef(voiceInputSettings.shortcut);
const voiceInputStateRef = useRef(voiceInput.state);
const voiceInputGetStateRef = useRef(voiceInput.getState);
+ const voiceInputPrepareRef = useRef(voiceInput.prepare);
const voiceInputStopRef = useRef(handleVoiceInputStopWithRefinement);
voiceInputBusyRef.current = voiceInput.isBusy;
voiceDraftTextRef.current = voiceInput.draftText;
@@ -3506,6 +3507,7 @@ export function ChatInput({
useEffect(() => {
voiceInputStateRef.current = voiceInput.state;
voiceInputGetStateRef.current = voiceInput.getState;
+ voiceInputPrepareRef.current = voiceInput.prepare;
voiceInputStopRef.current = handleVoiceInputStopWithRefinement;
voiceInputCancelRef.current = voiceInput.cancel;
handleVoiceInputStartRef.current = handleVoiceInputStart;
@@ -3520,6 +3522,7 @@ export function ChatInput({
handleVoiceInputStopWithRefinement,
voiceInput.cancel,
voiceInput.getState,
+ voiceInput.prepare,
voiceInput.state,
]);
@@ -3717,11 +3720,21 @@ export function ChatInput({
};
}, []);
- // 长按输入框语音输入(语音设置里打开,默认关闭):输入框不加任何提示,按住鼠标
+ // 长按输入框语音输入默认开启,手动配置可关闭。按住鼠标
// 左键不动片刻即开始录音,松开结束。按下由输入框的 onMouseDown 接入(要读
// detail 才能排除双击);移动/松开挂在 window 上。真正开始录音后再指针捕获,
// 拖出窗口松开也能停;等待期间不捕获,免得打断选字。
const composerLongPressVoiceInputEnabled = voiceInputSettings.composerLongPressEnabled;
+ const [composerVoiceHovered, setComposerVoiceHovered] = useState(false);
+ const [composerVoiceHoverHintReady, setComposerVoiceHoverHintReady] = useState(false);
+ useEffect(() => {
+ setComposerVoiceHoverHintReady(false);
+ if (!composerVoiceHovered) return;
+ const timer = setTimeout(() => setComposerVoiceHoverHintReady(true), 500);
+ return () => clearTimeout(timer);
+ }, [composerVoiceHovered]);
+ const [composerVoiceHolding, setComposerVoiceHolding] = useState(false);
+ const [voiceButtonHolding, setVoiceButtonHolding] = useState(false);
const composerLongPressGestureRef = useRef<ComposerLongPressVoiceGesture | null>(null);
const armComposerLongPressCaptureRef = useRef<((pointerId: number, target: Element) => void) | null>(
null,
@@ -3729,6 +3742,26 @@ export function ChatInput({
useEffect(() => {
if (!composerLongPressVoiceInputEnabled) return;
let capture: { pointerId: number; target: Element } | null = null;
+ let selectionLockTarget: HTMLElement | null = null;
+ const clearSelectionLock = () => {
+ selectionLockTarget?.removeAttribute('data-voice-holding');
+ selectionLockTarget = null;
+ };
+ // pointerup is followed by mouseup/click and asynchronous selectionchange.
+ // Keep protecting this gesture until the next user input, not a timeout
+ // that could expire before Chromium finishes dispatching those events.
+ const handleNextInput = () => {
+ if (!gesture.isHolding()) clearSelectionLock();
+ };
+ const preventReleaseClick = (event: MouseEvent) => {
+ if (!selectionLockTarget || gesture.isHolding()) return;
+ if (event.detail === 0) {
+ clearSelectionLock();
+ return;
+ }
+ event.preventDefault();
+ event.stopPropagation();
+ };
const handleLostPointerCapture = (event: Event) => {
if (!(event instanceof PointerEvent)) return;
@@ -3752,9 +3785,13 @@ export function ChatInput({
const gesture = createComposerLongPressVoiceGesture({
getState: () => voiceInputGetStateRef.current(),
+ prepare: () => voiceInputPrepareRef.current?.(),
start: () => handleVoiceInputStartRef.current(),
stop: () => voiceInputStopRef.current(),
onHoldStart: () => {
+ setComposerVoiceHolding(true);
+ selectionLockTarget = editorRef.current?.view.dom ?? null;
+ selectionLockTarget?.setAttribute('data-voice-holding', 'true');
if (!capture) return;
try {
capture.target.setPointerCapture(capture.pointerId);
@@ -3763,7 +3800,13 @@ export function ChatInput({
capture = null;
}
},
- onHoldEnd: releaseCapture,
+ onHoldEnd: () => {
+ setComposerVoiceHolding(false);
+ // Restore hit-testing for the next press immediately; only the
+ // selectionchange/click guard needs to outlive this release.
+ selectionLockTarget?.removeAttribute('data-voice-holding');
+ releaseCapture();
+ },
});
composerLongPressGestureRef.current = gesture;
armComposerLongPressCaptureRef.current = (pointerId, target) => {
@@ -3781,6 +3824,7 @@ export function ChatInput({
};
const handleLeftButtonRelease = (event: PointerEvent | MouseEvent) => {
if (event.button !== 0) return;
+ if (selectionLockTarget) event.preventDefault();
gesture.release();
// 没到时长就松开时不会走 onHoldEnd,但仍要清掉按下时记下的 pointerId。
releaseCapture();
@@ -3788,13 +3832,50 @@ export function ChatInput({
const handleAbort = () => {
gesture.release();
releaseCapture();
+ clearSelectionLock();
+ };
+ // pointermove.preventDefault() does not suppress Chromium's compatibility
+ // mouse events or native selection/drag defaults after mousedown.
+ const preventHeldSelection = (event: Event) => {
+ if (selectionLockTarget) event.preventDefault();
+ };
+ // mousedown has already started Chromium's selection gesture before the
+ // hold timer fires. Cancelling later move/selectstart events alone cannot
+ // guarantee it stops. Keep the DOM selection at the editor's current
+ // insertion range before ProseMirror observes native selection changes.
+ // Read the current state each time so streamed dictation can move it.
+ const preserveHeldSelection = (event: Event) => {
+ if (!selectionLockTarget) return;
+ const current = editorRef.current;
+ if (!current || current.isDestroyed) return;
+ const selection = current.view.dom.ownerDocument.getSelection();
+ if (!selection?.anchorNode || !current.view.dom.contains(selection.anchorNode)) return;
+ const anchor = current.view.domAtPos(current.state.selection.anchor);
+ const head = current.view.domAtPos(current.state.selection.head);
+ if (
+ selection.anchorNode === anchor.node && selection.anchorOffset === anchor.offset &&
+ selection.focusNode === head.node && selection.focusOffset === head.offset
+ ) return;
+ event.stopImmediatePropagation();
+ selection.setBaseAndExtent(anchor.node, anchor.offset, head.node, head.offset);
+ };
+ const handleDragStart = (event: Event) => {
+ if (gesture.isHolding()) event.preventDefault();
+ else handleAbort();
};
window.addEventListener('pointermove', handlePointerMove, true);
+ window.addEventListener('pointerdown', handleNextInput, true);
+ window.addEventListener('mousedown', handleNextInput, true);
+ window.addEventListener('keydown', handleNextInput, true);
+ window.addEventListener('click', preventReleaseClick, true);
+ window.addEventListener('mousemove', preventHeldSelection, true);
+ window.addEventListener('selectstart', preventHeldSelection, true);
+ document.addEventListener('selectionchange', preserveHeldSelection, true);
window.addEventListener('pointerup', handleLeftButtonRelease, true);
window.addEventListener('mouseup', handleLeftButtonRelease, true);
window.addEventListener('pointercancel', handleAbort, true);
// 按在已选中的文字上会进入原生拖拽,之后不再有 move / up。
- window.addEventListener('dragstart', handleAbort, true);
+ window.addEventListener('dragstart', handleDragStart, true);
window.addEventListener('blur', handleAbort);
return () => {
armComposerLongPressCaptureRef.current = null;
@@ -3803,17 +3884,31 @@ export function ChatInput({
}
gesture.dispose();
releaseCapture();
+ clearSelectionLock();
window.removeEventListener('pointermove', handlePointerMove, true);
+ window.removeEventListener('pointerdown', handleNextInput, true);
+ window.removeEventListener('mousedown', handleNextInput, true);
+ window.removeEventListener('keydown', handleNextInput, true);
+ window.removeEventListener('click', preventReleaseClick, true);
+ window.removeEventListener('mousemove', preventHeldSelection, true);
+ window.removeEventListener('selectstart', preventHeldSelection, true);
+ document.removeEventListener('selectionchange', preserveHeldSelection, true);
window.removeEventListener('pointerup', handleLeftButtonRelease, true);
window.removeEventListener('mouseup', handleLeftButtonRelease, true);
window.removeEventListener('pointercancel', handleAbort, true);
- window.removeEventListener('dragstart', handleAbort, true);
+ window.removeEventListener('dragstart', handleDragStart, true);
window.removeEventListener('blur', handleAbort);
};
- }, [composerLongPressVoiceInputEnabled]);
+ }, [composerLongPressVoiceInputEnabled, storageKey]);
const handleComposerLongPressPointerDown = useCallback((event: ReactPointerEvent<HTMLDivElement>) => {
if (event.button !== 0) return;
+ const current = editorRef.current;
+ if (!current || current.isDestroyed || !(event.target instanceof Element)) return;
+ if (
+ !current.view.dom.contains(event.target) &&
+ !isComposerBlankPointerTarget(event.target, event.currentTarget, current.view.dom, event)
+ ) return;
armComposerLongPressCaptureRef.current?.(event.pointerId, event.currentTarget);
}, []);
@@ -3828,7 +3923,11 @@ export function ChatInput({
if (!editorInstance || editorInstance.isDestroyed) return;
const editorElement = editorInstance.view.dom;
const target = event.target;
- if (!(target instanceof Element) || !editorElement.contains(target)) return;
+ if (!(target instanceof Element)) return;
+ if (
+ !editorElement.contains(target) &&
+ !isComposerBlankPointerTarget(target, event.currentTarget, editorElement, event)
+ ) return;
// 提及 / 粘贴文本 / 引用这类 chip 自带点击与拖拽行为,不在上面起手。
const atom = target.closest('[contenteditable="false"]');
if (atom && atom !== editorElement && editorElement.contains(atom)) return;
@@ -8899,10 +8998,24 @@ export function ChatInput({
});
// 推荐 overlay 的可见判据:开关开启 + 有推荐词 + 输入框空 + 无附件/浏览器评论/语音草稿 + 输入框未锁定。
// composerMutationLocked 涵盖 disabled、sendDispatchInFlight、当前输入框所属语音及远程只读/锁定状态。
- const showRecommendationOverlay = shouldShowComposerPromptRecommendation({
- ...recommendationComposerState,
- prompt: recommendedPrompt,
- });
+ const voiceHintEmpty =
+ !!editor && !editor.isDestroyed && composerDocIsEmpty(editor.state.doc) && !hasVoiceDraftText;
+ const voiceHint = voiceHintEmpty
+ ? voiceBusyOnCurrentComposer && voiceInput.state === 'listening'
+ ? t(
+ composerVoiceHolding || voiceButtonHolding
+ ? 'newChat.chatInput.voiceInput.speakRelease'
+ : 'newChat.chatInput.voiceInput.speakPressAgain',
+ )
+ : composerVoiceHovered && composerVoiceHoverHintReady && composerLongPressVoiceInputEnabled && !composerTypingLocked
+ ? t('newChat.chatInput.voiceInput.holdComposer')
+ : null
+ : null;
+ const showRecommendationOverlay =
+ !voiceHint && shouldShowComposerPromptRecommendation({
+ ...recommendationComposerState,
+ prompt: recommendedPrompt,
+ });
// handleKeyDown 的稳定闭包只按真实可见性接受 Tab,避免隐藏推荐被误填入。
showRecommendationRef.current = showRecommendationOverlay;
// 首页建议悬停预览:输入框锁定(发送中 / 语音占用 / 禁用)时不预览,免得遮住进行中的状态。
@@ -8911,7 +9024,7 @@ export function ChatInput({
const [dismissedPreviewPrompt, setDismissedPreviewPrompt] = useState<string | null>(null);
if (!previewPrompt && dismissedPreviewPrompt !== null) setDismissedPreviewPrompt(null);
const showPromptPreview =
- !!previewPrompt && previewPrompt !== dismissedPreviewPrompt && !composerMutationLocked;
+ !!previewPrompt && previewPrompt !== dismissedPreviewPrompt && !composerMutationLocked && !voiceHint;
promptPreviewInputGuardRef.current = () => {
if (!showPromptPreview) return false;
setDismissedPreviewPrompt(previewPrompt ?? null);
@@ -9176,12 +9289,22 @@ export function ChatInput({
],
)}
data-split-group-composer-drop-target
+ onPointerDown={handleComposerLongPressPointerDown}
+ onPointerOver={(event) => {
+ const editorDom = editor && !editor.isDestroyed ? editor.view.dom : null;
+ setComposerVoiceHovered(
+ (event.target instanceof Element && !!editorDom?.contains(event.target)) ||
+ isComposerBlankPointerTarget(event.target, event.currentTarget, editorDom, event),
+ );
+ }}
+ onPointerLeave={() => setComposerVoiceHovered(false)}
// 卡片里的空白(文字行下方的空隙、工具栏两组按钮之间的空档、四周
// padding)没有元素承接点击:浏览器默认会把焦点从 contenteditable 撤到
// <body>,正在输入的光标凭空消失;而点空白又该能进入输入态。所以先
// preventDefault 掉这次默认的焦点转移,再按需补一次不带坐标的 focus ——
// 「进入输入态」归空白区,「定位插入点」只归点击文字那一行。
onMouseDown={(event) => {
+ handleComposerLongPressMouseDown(event);
if (event.button !== 0) return;
if (
!isComposerBlankPointerTarget(
@@ -9435,10 +9558,9 @@ export function ChatInput({
className="relative w-full"
// 推荐词生效时由 CSS 关掉原生 placeholder,避免两行字叠在一起。
data-recommendation-active={showRecommendationOverlay ? 'true' : undefined}
+ data-voice-hint-active={voiceHint ? 'true' : undefined}
// 建议预览生效时由 CSS 隐去编辑器正文。
data-prompt-preview-active={showPromptPreview ? 'true' : undefined}
- onPointerDown={handleComposerLongPressPointerDown}
- onMouseDown={handleComposerLongPressMouseDown}
>
<EditorContent
editor={editor}
@@ -9452,6 +9574,14 @@ export function ChatInput({
voiceBusyOnCurrentComposer && voiceInput.draftText ? 'true' : undefined
}
/>
+ {voiceHint && (
+ <div
+ data-testid="composer-voice-hint"
+ className="pointer-events-none absolute inset-x-0 top-0 py-[3px] pr-[11px] text-15 leading-[1.467] font-normal text-[var(--chat-input-placeholder-subtle)]"
+ >
+ {voiceHint}
+ </div>
+ )}
{/* 建议预览:排版与原生 placeholder 对齐,py-[3px] 同下方推荐词 overlay 的说明。 */}
{showPromptPreview && (
<div
@@ -9927,6 +10057,8 @@ export function ChatInput({
}
shortcutLabel={voiceInputShortcutLabel}
onStart={handleVoiceInputStart}
+ onHoldChange={setVoiceButtonHolding}
+ emptyComposer={voiceHintEmpty}
onStop={handleVoiceInputPlainStop}
onStopAndSend={handleClickSend}
sendTargetRef={sendButtonRef}
@@ -10169,6 +10301,8 @@ export function ChatInput({
function VoiceInputButton({
state,
+ onHoldChange,
+ emptyComposer,
disabled,
shortcutLabel,
onStart,
@@ -10181,6 +10315,8 @@ function VoiceInputButton({
className,
}: {
state: import('@cindy/voice-input-core').VoiceInputState;
+ onHoldChange: (holding: boolean) => void;
+ emptyComposer: boolean;
disabled: boolean;
shortcutLabel: string;
onStart: () => boolean | Promise<boolean | void>;
@@ -10195,6 +10331,10 @@ function VoiceInputButton({
}) {
const { t } = useTranslation();
const [longPressActive, setLongPressActive] = useState(false);
+ useEffect(() => {
+ onHoldChange(longPressActive);
+ return () => onHoldChange(false);
+ }, [longPressActive, onHoldChange]);
const timerRef = useRef<number | null>(null);
const pointerIdRef = useRef<number | null>(null);
const lastPointerPointRef = useRef<{ clientX: number; clientY: number } | null>(null);
@@ -10259,8 +10399,16 @@ function VoiceInputButton({
} else if (activeRecording) {
label = t('newChat.chatInput.voiceInput.stop');
}
+ const hintLabel =
+ emptyComposer && activeRecording
+ ? t(
+ longPressActive
+ ? 'newChat.chatInput.voiceInput.speakRelease'
+ : 'newChat.chatInput.voiceInput.speakPressAgain',
+ )
+ : label;
const tooltipText =
- shortcutLabel && !longPressActive && !busy ? `${label} · ${shortcutLabel}` : label;
+ shortcutLabel && !activeRecording && !busy ? `${hintLabel} · ${shortcutLabel}` : hintLabel;
const controlledTooltipOpen = busy
? showProcessing
: longPressActive
@@ -10388,6 +10536,9 @@ function VoiceInputButton({
aria-pressed={activeRecording}
onPointerDown={(event) => {
if (event.button !== 0) return;
+ // A fresh press is never the synthetic click from the previous hold.
+ // Chromium may omit that click when the recording pill changes size.
+ clearSuppressNextClick();
event.preventDefault();
event.currentTarget.blur();
if (disabledOrBusy || listening) return;
diff --git a/apps/desktop/src/renderer/components/new-chat/__tests__/chatInputAnnotationAbort.test.tsx b/apps/desktop/src/renderer/components/new-chat/__tests__/chatInputAnnotationAbort.test.tsx
index f44528feb7..87d9f057b2 100644
--- a/apps/desktop/src/renderer/components/new-chat/__tests__/chatInputAnnotationAbort.test.tsx
+++ b/apps/desktop/src/renderer/components/new-chat/__tests__/chatInputAnnotationAbort.test.tsx
@@ -35,6 +35,7 @@ const h = vi.hoisted(() => ({ t: (key: string) => key, confirm: vi.fn(), editor:
draftSource: null as UseVoiceInputResult['draftSource'],
draftRange: null as UseVoiceInputResult['draftRange'],
cancel: vi.fn().mockResolvedValue(undefined),
+ prepare: vi.fn(),
realModelSelector: false,
listAgentSkills: vi.fn(),
localProviders: [] as ProviderView[],
@@ -88,7 +89,7 @@ vi.mock('@/voice-input/useVoiceInput', () => ({ useVoiceInput: (editor: Editor |
if (editor && editor !== h.editor) editor.on('transaction', ({ transaction }) => { if (transaction.docChanged) h.documentChanged(); });
h.editor = editor;
const state = h.voiceState ?? (h.listening ? 'listening' : 'idle');
- return { state, isListening: state === 'listening', isBusy: state === 'listening' || state === 'submitting' || state === 'refining', draftText: h.draftText, draftSource: h.draftSource, draftRange: h.draftRange, start: vi.fn(), stop: h.stop, cancel: h.cancel };
+ return { state, prepare: h.prepare, getState: () => h.voiceState ?? (h.listening ? 'listening' : 'idle'), isListening: state === 'listening', isBusy: state === 'listening' || state === 'submitting' || state === 'refining', draftText: h.draftText, draftSource: h.draftSource, draftRange: h.draftRange, start: () => { h.listening = true; return true; }, stop: h.stop, cancel: h.cancel };
} }));
vi.mock('@/hooks/useProviders', () => ({ useProviders: () => ({ providers: h.localProviders, providerOrder: [], loading: false }) }));
vi.mock('@/hooks/useDeviceProviders', () => ({ useDeviceProviders: () => ({ providers: [], loading: false, unsupported: false }) }));
@@ -139,6 +140,7 @@ beforeEach(() => {
h.draftSource = null;
h.draftRange = null;
h.stop.mockClear();
+ h.prepare.mockClear();
h.realModelSelector = false;
h.cancel.mockReset().mockResolvedValue(undefined);
h.atPanel = null;
@@ -233,6 +235,75 @@ it('locks editor and attachments for the whole voice run, but keeps permissions
}
});
+// User request 2026-10-24: empty composer hints must follow hover/hold/click,
+// and native selection must not take over a recording. Existing voice tests
+// only cover editability; minimal inputs below drive real DOM handlers (unit/jsdom).
+it('shows voice guidance only in an empty composer and distinguishes holding from clicking', async () => {
+ const view = renderComposer();
+ await waitFor(() => expect(h.editor).not.toBeNull());
+ const area = view.container.querySelector('[data-recommendation-active]')
+ ?? h.editor!.view.dom.parentElement!.parentElement!;
+ vi.useFakeTimers();
+ try {
+ fireEvent.pointerEnter(area);
+ await act(async () => { await vi.advanceTimersByTimeAsync(499); });
+ expect(screen.queryByTestId('composer-voice-hint')).toBeNull();
+ await act(async () => { await vi.advanceTimersByTimeAsync(1); });
+ expect(screen.getByTestId('composer-voice-hint').textContent).toBe('newChat.chatInput.voiceInput.holdComposer');
+ fireEvent.pointerLeave(area);
+ expect(screen.queryByTestId('composer-voice-hint')).toBeNull();
+ fireEvent.pointerEnter(area);
+ await act(async () => { await vi.advanceTimersByTimeAsync(250); });
+ fireEvent.pointerLeave(area);
+ await act(async () => { await vi.advanceTimersByTimeAsync(500); });
+ expect(screen.queryByTestId('composer-voice-hint')).toBeNull();
+ // jsdom has no hit-testing; returning null leaves ProseMirror's ordinary
+ // mousedown path inert while the real composer handler still receives it.
+ Object.defineProperty(document, 'elementFromPoint', { configurable: true, value: () => null });
+ fireEvent.mouseDown(h.editor!.view.dom, { button: 0, detail: 1, clientX: 10, clientY: 10 });
+ expect(h.prepare).toHaveBeenCalledOnce();
+ expect(h.listening).toBe(false);
+ await act(async () => { await vi.advanceTimersByTimeAsync(600); });
+ view.rerender(<ChatInput {...view.props} />);
+ expect(screen.getByTestId('composer-voice-hint').textContent).toBe('newChat.chatInput.voiceInput.speakRelease');
+ for (const type of ['selectstart', 'dragstart', 'mousemove']) {
+ const event = new Event(type, { bubbles: true, cancelable: true });
+ h.editor!.view.dom.dispatchEvent(event);
+ expect(event.defaultPrevented, type).toBe(true);
+ }
+ await act(async () => { fireEvent.mouseUp(window, { button: 0 }); });
+ expect(h.stop).toHaveBeenCalledOnce();
+ const selection = new Event('selectstart', { bubbles: true, cancelable: true });
+ h.editor!.view.dom.dispatchEvent(selection);
+ expect(selection.defaultPrevented).toBe(true);
+ const releaseClick = new MouseEvent('click', { bubbles: true, cancelable: true, detail: 1 });
+ h.editor!.view.dom.dispatchEvent(releaseClick);
+ expect(releaseClick.defaultPrevented).toBe(true);
+ fireEvent.pointerDown(document.body, { button: 0 });
+ const nextSelection = new Event('selectstart', { bubbles: true, cancelable: true });
+ h.editor!.view.dom.dispatchEvent(nextSelection);
+ expect(nextSelection.defaultPrevented).toBe(false);
+ } finally {
+ Reflect.deleteProperty(document, 'elementFromPoint');
+ vi.useRealTimers();
+ }
+
+ h.listening = false;
+ view.rerender(<ChatInput {...view.props} />);
+ fireEvent.click(screen.getByRole('button', { name: 'newChat.chatInput.voiceInput.start' }));
+ view.rerender(<ChatInput {...view.props} />);
+ expect(screen.getByTestId('composer-voice-hint').textContent).toBe('newChat.chatInput.voiceInput.speakPressAgain');
+ h.draftText = 'hello';
+ view.rerender(<ChatInput {...view.props} />);
+ expect(screen.queryByTestId('composer-voice-hint')).toBeNull();
+ h.draftText = '';
+ h.listening = false;
+ await act(async () => { h.editor!.commands.setContent('<p>Existing text</p>'); });
+ view.rerender(<ChatInput {...view.props} />);
+ fireEvent.pointerEnter(area);
+ expect(screen.queryByTestId('composer-voice-hint')).toBeNull();
+});
+
// #374 / voiceInputButtonAnchor: the real toolbar must keep the voice control
// immediately beside the main action while listening, submitting, refining,
// and returning to idle. A source-text guard could miss a misplaced rendered
diff --git a/apps/desktop/src/renderer/components/settings/VoiceInputSection.settings-search.ts b/apps/desktop/src/renderer/components/settings/VoiceInputSection.settings-search.ts
index 14e48c60ca..8d13b1fbdd 100644
--- a/apps/desktop/src/renderer/components/settings/VoiceInputSection.settings-search.ts
+++ b/apps/desktop/src/renderer/components/settings/VoiceInputSection.settings-search.ts
@@ -16,7 +16,6 @@ export default {
{ id: 'settings.voiceInput.history.label', fallbackTargetId: 'settings-panel-voice-input', tab: 'voice-input', targetId: 'settings-search-settings-voiceInput-history-label', titleKey: 'settings.voiceInput.history.label', sectionKey: 'settings.tabs.voiceInput' },
{ id: 'settings.voiceInput.sections.usageData', fallbackTargetId: 'settings-panel-voice-input', tab: 'voice-input', targetId: 'settings-search-voice-usage-data', titleKey: 'settings.voiceInput.sections.usageData', sectionKey: 'settings.tabs.voiceInput' },
{ id: 'settings.voiceInput.fastActivation.label', fallbackTargetId: 'settings-panel-voice-input', tab: 'voice-input', targetId: 'settings-search-settings-voiceInput-fastActivation-label', titleKey: 'settings.voiceInput.fastActivation.label', sectionKey: 'settings.tabs.voiceInput', descriptionKey: 'settings.voiceInput.fastActivation.hint' },
- { id: 'settings.voiceInput.composerLongPress.label', fallbackTargetId: 'settings-panel-voice-input', tab: 'voice-input', targetId: 'settings-search-settings-voiceInput-composerLongPress-label', titleKey: 'settings.voiceInput.composerLongPress.label', sectionKey: 'settings.tabs.voiceInput', descriptionKey: 'settings.voiceInput.composerLongPress.hint' },
{ id: 'settings.voiceInput.speakerFilter.label', fallbackTargetId: 'settings-panel-voice-input', tab: 'voice-input', targetId: 'settings-search-settings-voiceInput-speakerFilter-label', titleKey: 'settings.voiceInput.speakerFilter.label', sectionKey: 'settings.tabs.voiceInput', descriptionKey: 'settings.voiceInput.speakerFilter.hint', isVisible: ({ platform }) => platform === 'darwin' || platform === 'win32' },
{ id: 'settings.voiceInput.shortcut.label', fallbackTargetId: 'settings-panel-voice-input', tab: 'voice-input', targetId: 'settings-search-settings-voiceInput-shortcut-label', titleKey: 'settings.voiceInput.shortcut.label', sectionKey: 'settings.tabs.voiceInput', descriptionKey: 'settings.voiceInput.shortcut.hint' },
diff --git a/apps/desktop/src/renderer/components/settings/VoiceInputSection.tsx b/apps/desktop/src/renderer/components/settings/VoiceInputSection.tsx
index 910d758cdb..56f200d9f0 100644
--- a/apps/desktop/src/renderer/components/settings/VoiceInputSection.tsx
+++ b/apps/desktop/src/renderer/components/settings/VoiceInputSection.tsx
@@ -1038,8 +1038,6 @@ export function VoiceInputSection() {
setMuteSystemAudio,
setPlayInteractionSound,
setFastActivationEnabled,
- setComposerLongPressEnabled,
- resetComposerLongPressEnabled,
setRefinementEnabled,
setRefinementInstructions,
setAutoDictionaryEnabled,
@@ -2695,36 +2693,6 @@ export function VoiceInputSection() {
/>
</div>
- <div className="flex items-center justify-between gap-5 border-t border-[var(--settings-theme-card-border)] pt-4">
- <div className="flex min-w-0 flex-col gap-1">
- <p id="settings-search-settings-voiceInput-composerLongPress-label"
- className="text-13 font-medium text-[var(--settings-section-title)]"
- style={{ letterSpacing: '0.12px' }}
- >
- {t('settings.voiceInput.composerLongPress.label')}
- </p>
- <p className="text-12 leading-[1.4] text-[var(--text-secondary)]">
- {t('settings.voiceInput.composerLongPress.hint')}
- </p>
- </div>
-
- <div className="flex shrink-0 items-center gap-2">
- <DefaultOverrideControls
- isCustomized={settings.composerLongPressEnabledOverride != null}
- onReset={() => {
- void resetComposerLongPressEnabled().then((ok) => {
- if (ok) toast.success(t('settings.defaults.restored'));
- });
- }}
- />
- <Switch
- checked={settings.composerLongPressEnabled}
- onCheckedChange={setComposerLongPressEnabled}
- aria-label={t('settings.voiceInput.composerLongPress.ariaLabel')}
- />
- </div>
- </div>
-
<div className="flex items-center justify-between gap-5 border-t border-[var(--settings-theme-card-border)] pt-4">
<div className="flex min-w-0 flex-col gap-1">
<p id="settings-search-settings-voiceInput-interactionSound-label"
diff --git a/apps/desktop/src/renderer/hooks/__tests__/useVoiceInputSettings.overrideReset.test.ts b/apps/desktop/src/renderer/hooks/__tests__/useVoiceInputSettings.overrideReset.test.ts
index b02f29d468..e5ed6f8b23 100644
--- a/apps/desktop/src/renderer/hooks/__tests__/useVoiceInputSettings.overrideReset.test.ts
+++ b/apps/desktop/src/renderer/hooks/__tests__/useVoiceInputSettings.overrideReset.test.ts
@@ -62,7 +62,7 @@ describe('voice input override restore defaults', () => {
expect(patch).toEqual({ composerLongPressEnabled: null });
const next = { ...settings };
delete next.composerLongPressEnabledOverride;
- next.composerLongPressEnabled = false;
+ next.composerLongPressEnabled = getDefaultVoiceInputSettings('darwin').composerLongPressEnabled;
return next;
});
@@ -74,7 +74,7 @@ describe('voice input override restore defaults', () => {
expect(ok).toBe(true);
expect(updateSettings).toHaveBeenCalledTimes(1);
expect(updateSettings).toHaveBeenCalledWith({ composerLongPressEnabled: null });
- expect(result.current.settings.composerLongPressEnabled).toBe(false);
+ expect(result.current.settings.composerLongPressEnabled).toBe(true);
expect(result.current.settings.composerLongPressEnabledOverride).toBeUndefined();
});
diff --git a/apps/desktop/src/renderer/i18n/locales/en/common.json b/apps/desktop/src/renderer/i18n/locales/en/common.json
index 1bc434871e..11b9193578 100644
--- a/apps/desktop/src/renderer/i18n/locales/en/common.json
+++ b/apps/desktop/src/renderer/i18n/locales/en/common.json
@@ -11179,6 +11179,9 @@
"start": "Start voice input",
"stop": "Stop voice input",
"releaseToStop": "Recording. Release to stop.",
+ "holdComposer": "Hold the input box to dictate",
+ "speakRelease": "Start speaking. Release to finish.",
+ "speakPressAgain": "Start speaking. Press again to finish.",
"releaseToSend": "Release to send",
"finishAndSend": "Finish voice input and send",
"refining": "Refining",
diff --git a/apps/desktop/src/renderer/i18n/locales/ja/common.json b/apps/desktop/src/renderer/i18n/locales/ja/common.json
index 48222ea6ca..e2d2642288 100644
--- a/apps/desktop/src/renderer/i18n/locales/ja/common.json
+++ b/apps/desktop/src/renderer/i18n/locales/ja/common.json
@@ -11144,6 +11144,9 @@
"start": "音声入力を開始",
"stop": "音声入力を停止",
"releaseToStop": "音声入力中、離すと終了",
+ "holdComposer": "入力欄を長押しして音声入力",
+ "speakRelease": "話してください。離すと終了",
+ "speakPressAgain": "話してください。もう一度押すと終了",
"releaseToSend": "離すと送信",
"finishAndSend": "音声入力を終了して送信",
"refining": "整えています",
diff --git a/apps/desktop/src/renderer/i18n/locales/ko/common.json b/apps/desktop/src/renderer/i18n/locales/ko/common.json
index 72ef531752..f603620879 100644
--- a/apps/desktop/src/renderer/i18n/locales/ko/common.json
+++ b/apps/desktop/src/renderer/i18n/locales/ko/common.json
@@ -11144,6 +11144,9 @@
"start": "음성 입력 시작",
"stop": "음성 입력 중지",
"releaseToStop": "음성 입력 중, 놓으면 종료",
+ "holdComposer": "입력란을 길게 눌러 음성 입력",
+ "speakRelease": "말씀하세요. 놓으면 종료",
+ "speakPressAgain": "말씀하세요. 다시 누르면 종료",
"releaseToSend": "놓으면 전송",
"finishAndSend": "음성 입력을 끝내고 전송",
"refining": "다듬는 중",
diff --git a/apps/desktop/src/renderer/i18n/locales/zh-CN/common.json b/apps/desktop/src/renderer/i18n/locales/zh-CN/common.json
index 793d1ea4ca..5e461dec39 100644
--- a/apps/desktop/src/renderer/i18n/locales/zh-CN/common.json
+++ b/apps/desktop/src/renderer/i18n/locales/zh-CN/common.json
@@ -11144,6 +11144,9 @@
"start": "开始语音输入",
"stop": "停止语音输入",
"releaseToStop": "语音输入中,松开结束",
+ "holdComposer": "长按输入框,语音输入",
+ "speakRelease": "开始说话,松开结束",
+ "speakPressAgain": "开始说话,再次按下结束",
"releaseToSend": "松开发送",
"finishAndSend": "结束语音输入并发送",
"refining": "正在润色",
diff --git a/apps/desktop/src/renderer/i18n/locales/zh-TW/common.json b/apps/desktop/src/renderer/i18n/locales/zh-TW/common.json
index 5ca5557bd5..24eb90eda3 100644
--- a/apps/desktop/src/renderer/i18n/locales/zh-TW/common.json
+++ b/apps/desktop/src/renderer/i18n/locales/zh-TW/common.json
@@ -11144,6 +11144,9 @@
"start": "開始語音輸入",
"stop": "停止語音輸入",
"releaseToStop": "語音輸入中,鬆開結束",
+ "holdComposer": "長按輸入框,語音輸入",
+ "speakRelease": "開始說話,鬆開結束",
+ "speakPressAgain": "開始說話,再按一下結束",
"releaseToSend": "鬆開傳送",
"finishAndSend": "結束語音輸入並傳送",
"refining": "正在潤色",
diff --git a/apps/desktop/src/renderer/styles/globals.css b/apps/desktop/src/renderer/styles/globals.css
index 277cb89ee5..b037de5dc5 100644
--- a/apps/desktop/src/renderer/styles/globals.css
+++ b/apps/desktop/src/renderer/styles/globals.css
@@ -339,10 +339,21 @@ html:lang(ko) {
content: none;
}
/* 输入框推荐提示词生效时,推荐词由 ChatInput 的 overlay 承载,关掉原生 placeholder。 */
+[data-voice-hint-active='true'] .ProseMirror > p.is-editor-empty:first-child::before,
+[data-voice-hint-active='true'] .ProseMirror > p.is-empty:first-child:only-child::before,
[data-recommendation-active='true'] .ProseMirror > p.is-editor-empty:first-child::before,
[data-recommendation-active='true'] .ProseMirror > p.is-empty:first-child:only-child::before {
content: none;
}
+.ProseMirror[data-voice-holding='true'],
+.ProseMirror[data-voice-holding='true'] * {
+ -webkit-user-select: none;
+ user-select: none;
+}
+/* 空框用完整语音提示替代语音光标,避免图标压住提示首字。 */
+[data-voice-hint-active='true'] [data-voice-caret] {
+ visibility: hidden;
+}
/* 首页建议悬停预览:预览由 ChatInput 的 overlay 承载,暂时隐去正文/placeholder;
预览不改变编辑器高度,见 ChatInput 的 promptPreviewLines。 */
[data-prompt-preview-active='true'] .ProseMirror {
diff --git a/apps/desktop/src/renderer/voice-input/WebMicAudioEngine.ts b/apps/desktop/src/renderer/voice-input/WebMicAudioEngine.ts
index f05be36aed..b3eb9cf052 100644
--- a/apps/desktop/src/renderer/voice-input/WebMicAudioEngine.ts
+++ b/apps/desktop/src/renderer/voice-input/WebMicAudioEngine.ts
@@ -21,6 +21,8 @@ export type WebMicAudioEngineOptions = {
audioProcessing?: boolean | AudioProcessingConfig;
latencyMs?: number;
keepAlive?: boolean;
+ /** Borrow a prepared session for this recording only, including device fallback. */
+ releaseAfterUse?: boolean;
onStateChange?: (event: string, details?: Record<string, unknown>) => void;
onInterrupted?: (message: string) => void;
};
@@ -853,6 +855,40 @@ async function getOrCreateKeepAliveSession(
return keepAliveSessionPromise;
}
+/** Prepare an inactive microphone for one press; never activate PCM or replace another owner's device. */
+export function prepareVoiceInputMicrophone(options: WebMicAudioEngineOptions): (() => void) | undefined {
+ const normalized = normalizeKeepAliveOptions(options);
+ // A failed startup can leave its settling promise without a session. Let
+ // normal recording recover it; preparation must own a session synchronously.
+ if (keepAliveSessionPromise && !keepAliveSession) return;
+ if (keepAliveSession && (
+ keepAliveSession.isBusy() || keepAliveSession.isStaleAfterDeviceChange() ||
+ keepAliveSession.key !== keepAliveKey(normalized)
+ )) return;
+ // With no replacement needed, getOrCreate publishes the session before its
+ // first await. Retain that identity so a late cancellation cannot close a
+ // different session, including while getUserMedia is still pending.
+ const ready = getOrCreateKeepAliveSession(normalized);
+ const session = keepAliveSession;
+ void ready.then((prepared) => {
+ if (options.keepAlive && !prepared.isBusy()) {
+ scheduleKeepAliveIdleDispose(prepared, 'idle_timeout_after_prewarm', { refresh: false });
+ }
+ }).catch(() => undefined);
+ let released = false;
+ return () => {
+ if (released) return;
+ released = true;
+ if (options.keepAlive || !session || keepAliveSession !== session) return;
+ if (session.isBusy()) {
+ // Recording adopted it. Let its normal stop close the temporary device.
+ session.markStaleForReplacement();
+ } else {
+ void disposeKeepAliveVoiceInputMicrophone('press_preparation_cancelled').catch(() => undefined);
+ }
+ };
+}
+
export async function prewarmVoiceInputMicrophone(options: WebMicAudioEngineOptions): Promise<void> {
const normalized = normalizeKeepAliveOptions(options);
// Prewarm can land mid-dictation: another ChatInput mounts, or a settings
@@ -1019,6 +1055,7 @@ export class WebMicAudioEngine {
private readonly audioProcessing: boolean | AudioProcessingConfig;
private readonly latencyMs?: number;
private readonly keepAlive: boolean;
+ private readonly releaseAfterUse: boolean;
private readonly onStateChange?: (event: string, details?: Record<string, unknown>) => void;
private readonly onInterrupted?: (message: string) => void;
private keepAliveSession?: KeepAliveMicSession;
@@ -1053,6 +1090,7 @@ export class WebMicAudioEngine {
this.audioProcessing = options.audioProcessing ?? true;
this.latencyMs = options.latencyMs;
this.keepAlive = options.keepAlive ?? false;
+ this.releaseAfterUse = options.releaseAfterUse ?? false;
this.onStateChange = options.onStateChange;
this.onInterrupted = options.onInterrupted;
}
@@ -1309,6 +1347,7 @@ export class WebMicAudioEngine {
this.clearWatchdog();
if (this.usingKeepAliveSession) {
const session = this.keepAliveSession;
+ if (this.releaseAfterUse) session?.markStaleForReplacement();
// Drop the claim taken in startKeepAlive, then only tear the session down
// if nobody else still holds one. deactivate() clears the shared PCM
// callback, sends setActive=false and detaches the output path — doing
diff --git a/apps/desktop/src/renderer/voice-input/__tests__/VoiceInputSection.overrideReset.test.tsx b/apps/desktop/src/renderer/voice-input/__tests__/VoiceInputSection.overrideReset.test.tsx
index 2fe0c28eb1..f91992e3b4 100644
--- a/apps/desktop/src/renderer/voice-input/__tests__/VoiceInputSection.overrideReset.test.tsx
+++ b/apps/desktop/src/renderer/voice-input/__tests__/VoiceInputSection.overrideReset.test.tsx
@@ -90,8 +90,6 @@ it('keeps custom ASR URL validation inline and associated through repeated edits
});
const cases = [
- { field: 'composerLongPressEnabled', override: 'composerLongPressEnabledOverride',
- label: 'settings.voiceInput.composerLongPress.ariaLabel' },
{ field: 'dictionarySyncEnabled', override: 'dictionarySyncEnabledOverride',
label: 'settings.voiceInput.refinement.dictionary.deviceSync.ariaLabel' },
] as const;
@@ -103,7 +101,8 @@ const cases = [
* clicked this component, so removing its controls or disconnecting their props escaped coverage.
* Minimal input: one explicit override opposite to its default (including false for dictionary sync).
* Unit/jsdom: a separate file is needed because recordingGate is a node-only source guard; no
- * VoiceInputSection component harness exists to extend. Both controls share this single fixture.
+ * VoiceInputSection component harness exists to extend. The 2026-10-24 product decision hides
+ * long-press dictation even with an explicit override; dictionary sync keeps its restore control.
* Measured alone: 3.25s wall / 2.19s Vitest (1.11s collect, 550ms jsdom, 281ms tests).
*/
describe('VoiceInputSection restore defaults', () => {
@@ -130,9 +129,8 @@ describe('VoiceInputSection restore defaults', () => {
expect(row().queryByText('settings.defaults.customizedBadge')).toBeNull();
expect(toast.success).toHaveBeenCalledExactlyOnceWith('settings.defaults.restored');
- const other = cases.find(candidate => candidate.field !== field)!;
- const otherToggle = screen.getByRole('switch', { name: other.label });
- expect(otherToggle.getAttribute('aria-checked')).toBe(String(!defaults[other.field]));
- expect(within(otherToggle.parentElement!.parentElement!).getByRole('button', { name: restoreLabel })).toBeTruthy();
+ // Long-press dictation is now a hidden override, even when explicitly configured.
+ expect(screen.queryByRole('switch', { name: 'settings.voiceInput.composerLongPress.ariaLabel' })).toBeNull();
+ expect(settings.composerLongPressEnabledOverride).toBe(false);
});
});
diff --git a/apps/desktop/src/renderer/voice-input/__tests__/WebMicAudioEngine.prewarm.test.ts b/apps/desktop/src/renderer/voice-input/__tests__/WebMicAudioEngine.prewarm.test.ts
index af004f6b28..9ce80b1979 100644
--- a/apps/desktop/src/renderer/voice-input/__tests__/WebMicAudioEngine.prewarm.test.ts
+++ b/apps/desktop/src/renderer/voice-input/__tests__/WebMicAudioEngine.prewarm.test.ts
@@ -5,6 +5,7 @@ import {
disposeKeepAliveVoiceInputMicrophone,
isMicrophoneDeviceUnavailableError,
prewarmVoiceInputMicrophoneWithAutomaticFallback,
+ prepareVoiceInputMicrophone,
WebMicAudioEngine,
} from '../WebMicAudioEngine';
@@ -114,6 +115,75 @@ describe('prewarmVoiceInputMicrophoneWithAutomaticFallback', () => {
vi.restoreAllMocks();
});
+ it('prepares without activating PCM and closes a cancelled press', async () => {
+ const release = prepareVoiceInputMicrophone({ workletUrl: 'https://app.local/pcm16k-worklet.js' });
+ await vi.waitFor(() => expect(AudioWorkletNode).toHaveBeenCalledOnce());
+ const node = vi.mocked(AudioWorkletNode).mock.results[0].value;
+ expect(node.port.postMessage).toHaveBeenCalledWith({ type: 'setActive', active: false, reset: true });
+ expect(node.port.postMessage).not.toHaveBeenCalledWith(expect.objectContaining({ active: true }));
+ release?.();
+ await vi.waitFor(() => expect(track.stopped).toBe(true));
+ });
+
+ it('reuses press preparation for capture and closes it after recording', async () => {
+ const options = { workletUrl: 'https://app.local/pcm16k-worklet.js' };
+ const release = prepareVoiceInputMicrophone(options);
+ const engine = new WebMicAudioEngine({ ...options, keepAlive: true });
+ const started = engine.start();
+ release?.();
+ await started;
+ expect(getUserMedia).toHaveBeenCalledOnce();
+ expect(track.stopped).toBe(false);
+ expect(prepareVoiceInputMicrophone(options)).toBeUndefined();
+ await engine.stop();
+ expect(track.stopped).toBe(true);
+ });
+
+ it('closes a late microphone when the press was already cancelled', async () => {
+ let finish!: (stream: unknown) => void;
+ getUserMedia.mockImplementation(() => new Promise((resolve) => { finish = resolve; }));
+ const release = prepareVoiceInputMicrophone({ workletUrl: 'https://app.local/pcm16k-worklet.js' });
+ await vi.waitFor(() => expect(getUserMedia).toHaveBeenCalledOnce());
+ release?.();
+ finish({ getAudioTracks: () => [track], getTracks: () => [track] });
+ await vi.waitFor(() => expect(track.stopped).toBe(true));
+ expect(AudioWorkletNode).not.toHaveBeenCalled();
+ });
+
+ it('closes a temporary replacement device even when preparation could not open the selected one', async () => {
+ const release = prepareVoiceInputMicrophone({
+ workletUrl: 'https://app.local/pcm16k-worklet.js', deviceId: 'missing-device',
+ });
+ await vi.waitFor(() => expect(enumerateDevices).toHaveBeenCalledOnce());
+ release?.();
+ const engine = new WebMicAudioEngine({
+ workletUrl: 'https://app.local/pcm16k-worklet.js', keepAlive: true, releaseAfterUse: true,
+ });
+ await engine.start();
+ expect(track.stopped).toBe(false);
+ await engine.stop();
+ expect(track.stopped).toBe(true);
+ });
+
+ it('does not reopen a failed device after the next preparation was cancelled', async () => {
+ let rejectDevice!: (error: Error) => void;
+ getUserMedia.mockImplementationOnce(() => new Promise((_resolve, reject) => { rejectDevice = reject; }));
+ const options = { workletUrl: 'https://app.local/pcm16k-worklet.js' };
+ const release = prepareVoiceInputMicrophone(options);
+ await vi.waitFor(() => expect(getUserMedia).toHaveBeenCalledOnce());
+ rejectDevice(new DOMException('Device unavailable', 'NotReadableError'));
+ await new Promise((resolve) => setTimeout(resolve, 0));
+ release?.();
+ prepareVoiceInputMicrophone(options)?.();
+ await new Promise((resolve) => setTimeout(resolve, 0));
+ expect(getUserMedia).toHaveBeenCalledOnce();
+ const engine = new WebMicAudioEngine({ ...options, keepAlive: true, releaseAfterUse: true });
+ await engine.start();
+ expect(getUserMedia).toHaveBeenCalledTimes(2);
+ await engine.stop();
+ expect(track.stopped).toBe(true);
+ });
+
it('requests microphone permission only once when keep-alive capture is denied', async () => {
getUserMedia.mockRejectedValue(new DOMException('Permission denied', 'NotAllowedError'));
const engine = new WebMicAudioEngine({
diff --git a/apps/desktop/src/renderer/voice-input/__tests__/composerLongPressGesture.test.ts b/apps/desktop/src/renderer/voice-input/__tests__/composerLongPressGesture.test.ts
index 298fc7b73d..7d917c313d 100644
--- a/apps/desktop/src/renderer/voice-input/__tests__/composerLongPressGesture.test.ts
+++ b/apps/desktop/src/renderer/voice-input/__tests__/composerLongPressGesture.test.ts
@@ -29,13 +29,18 @@ function setup(initialState: VoiceInputState = 'idle') {
const stop = vi.fn(async () => {
state = 'done';
});
+ const releasePreparation = vi.fn();
+ const prepare = vi.fn(() => releasePreparation);
const gesture = createComposerLongPressVoiceGesture({
getState: () => state,
+ prepare,
start,
stop,
});
return {
gesture,
+ prepare,
+ releasePreparation,
start,
stop,
getState: () => state,
@@ -55,16 +60,19 @@ describe('createComposerLongPressVoiceGesture', () => {
});
it('按住不动到时长才开始录音,松开结束', async () => {
- const { gesture, start, stop, getState } = setup();
+ const { gesture, start, stop, getState, prepare, releasePreparation } = setup();
gesture.press(origin);
+ expect(prepare).toHaveBeenCalledOnce();
vi.advanceTimersByTime(COMPOSER_LONG_PRESS_VOICE_INPUT_MS - 1);
await settle();
expect(start).not.toHaveBeenCalled();
+ expect(releasePreparation).not.toHaveBeenCalled();
vi.advanceTimersByTime(1);
await settle();
expect(start).toHaveBeenCalledOnce();
+ expect(releasePreparation).toHaveBeenCalledOnce();
expect(getState()).toBe('listening');
expect(gesture.isHolding()).toBe(true);
@@ -75,11 +83,12 @@ describe('createComposerLongPressVoiceGesture', () => {
});
it('没到时长就松开是普通点击,不录音', async () => {
- const { gesture, start, stop } = setup();
+ const { gesture, start, stop, releasePreparation } = setup();
gesture.press(origin);
vi.advanceTimersByTime(COMPOSER_LONG_PRESS_VOICE_INPUT_MS / 2);
gesture.release();
+ expect(releasePreparation).toHaveBeenCalledOnce();
vi.advanceTimersByTime(COMPOSER_LONG_PRESS_VOICE_INPUT_MS);
await settle();
@@ -88,11 +97,12 @@ describe('createComposerLongPressVoiceGesture', () => {
});
it('等待期间拖动超出容差视为选字,放弃长按', async () => {
- const { gesture, start } = setup();
+ const { gesture, start, releasePreparation } = setup();
gesture.press(origin);
gesture.move({ x: origin.x + COMPOSER_LONG_PRESS_MOVE_TOLERANCE_PX, y: origin.y });
gesture.move({ x: origin.x + COMPOSER_LONG_PRESS_MOVE_TOLERANCE_PX + 1, y: origin.y });
+ expect(releasePreparation).toHaveBeenCalledOnce();
vi.advanceTimersByTime(COMPOSER_LONG_PRESS_VOICE_INPUT_MS);
await settle();
diff --git a/apps/desktop/src/renderer/voice-input/__tests__/useVoiceInput.startup.test.ts b/apps/desktop/src/renderer/voice-input/__tests__/useVoiceInput.startup.test.ts
index 720fb17237..60da41b906 100644
--- a/apps/desktop/src/renderer/voice-input/__tests__/useVoiceInput.startup.test.ts
+++ b/apps/desktop/src/renderer/voice-input/__tests__/useVoiceInput.startup.test.ts
@@ -18,7 +18,9 @@ const mocks = vi.hoisted(() => ({
startEngine: vi.fn<() => Promise<void>>(),
stopEngine: vi.fn(),
drainEngine: vi.fn(),
- engines: [] as Array<{ emit: (chunk: PcmChunk) => void }>,
+ engines: [] as Array<{ emit: (chunk: PcmChunk) => void; keepAlive: boolean }>,
+ prepareMicrophone: vi.fn(),
+ releasePreparation: vi.fn(),
promptExpired: vi.fn(),
}));
vi.mock('react-i18next', () => ({ useTranslation: () => ({ t: mocks.translate }) }));
@@ -49,9 +51,10 @@ vi.mock('../audioContextPool', () => ({
vi.mock('../WebMicAudioEngine', () => ({
WebMicAudioEngine: class {
emit = (_chunk: PcmChunk) => {};
- constructor() {
+ constructor(readonly options: { keepAlive: boolean }) {
mocks.engines.push(this);
}
+ get keepAlive() { return this.options.keepAlive; }
onPcm16k(callback: (chunk: PcmChunk) => void) {
this.emit = callback;
}
@@ -74,6 +77,7 @@ vi.mock('../WebMicAudioEngine', () => ({
disposeKeepAliveVoiceInputMicrophone: vi.fn().mockResolvedValue(undefined),
prewarmVoiceInputBenchmarkFixture: vi.fn().mockResolvedValue(undefined),
prewarmVoiceInputMicrophoneWithAutomaticFallback: vi.fn().mockResolvedValue(undefined),
+ prepareVoiceInputMicrophone: mocks.prepareMicrophone,
}));
import { useVoiceInput } from '../useVoiceInput';
@@ -150,6 +154,8 @@ beforeEach(() => {
mocks.startEngine.mockReset().mockResolvedValue(undefined);
mocks.stopEngine.mockReset();
mocks.drainEngine.mockReset();
+ mocks.releasePreparation.mockReset();
+ mocks.prepareMicrophone.mockReset().mockReturnValue(mocks.releasePreparation);
});
afterEach(() => {
cleanup();
@@ -158,6 +164,31 @@ afterEach(() => {
});
describe('local voice capture independent of startup checks and cloud', () => {
+ it('prepares only authorized devices without claiming recording or starting ASR', async () => {
+ const h = mount();
+ h.result.current.prepare();
+ expect(mocks.prepareMicrophone).not.toHaveBeenCalled();
+ h.api.getMicrophonePermissionCached.mockReturnValue({ ok: true, status: 'granted' });
+ const release = h.result.current.prepare();
+ expect(mocks.prepareMicrophone).toHaveBeenCalledOnce();
+ expect(h.result.current.state).toBe('idle');
+ expect(h.api.start).not.toHaveBeenCalled();
+ expect(mocks.engines).toHaveLength(0);
+ await act(async () => { h.result.current.start(); });
+ expect(mocks.engines[0].keepAlive).toBe(true);
+ release?.();
+ expect(mocks.releasePreparation).toHaveBeenCalledOnce();
+ });
+
+ it('releases unused press preparation when the composer unmounts', () => {
+ const h = mount();
+ h.api.getMicrophonePermissionCached.mockReturnValue({ ok: true, status: 'granted' });
+ h.result.current.prepare();
+ h.unmount();
+ expect(mocks.releasePreparation).toHaveBeenCalledOnce();
+ expect(h.api.start).not.toHaveBeenCalled();
+ });
+
it('ends before the microphone produces its first frame and closes the late microphone', async () => {
const h = mount();
const device = deferred<void>();
diff --git a/apps/desktop/src/renderer/voice-input/captureSession.ts b/apps/desktop/src/renderer/voice-input/captureSession.ts
index 7d11f0048b..5297ed8c70 100644
--- a/apps/desktop/src/renderer/voice-input/captureSession.ts
+++ b/apps/desktop/src/renderer/voice-input/captureSession.ts
@@ -48,6 +48,7 @@ type VoiceInputCaptureSessionOptions = {
workletUrl: string;
deviceId?: string;
fastActivationEnabled: boolean;
+ temporaryMicrophone?: boolean;
getRunId: () => string | null;
setEngine: (engine: WebMicAudioEngine | null) => void;
isCurrentEngine?: (engine: WebMicAudioEngine) => boolean;
@@ -84,6 +85,7 @@ export async function startVoiceInputCaptureSession(
workletUrl: options.workletUrl,
deviceId,
...createVoiceInputAudioProfile(options.fastActivationEnabled),
+ releaseAfterUse: options.temporaryMicrophone,
onStateChange: options.onStateChange,
onInterrupted: options.onInterrupted,
});
diff --git a/apps/desktop/src/renderer/voice-input/composerLongPressGesture.ts b/apps/desktop/src/renderer/voice-input/composerLongPressGesture.ts
index e077680e54..0634eaf233 100644
--- a/apps/desktop/src/renderer/voice-input/composerLongPressGesture.ts
+++ b/apps/desktop/src/renderer/voice-input/composerLongPressGesture.ts
@@ -11,6 +11,8 @@ export type ComposerLongPressVoiceGestureOptions = {
holdMs?: number;
moveTolerancePx?: number;
getState: () => VoiceInputState;
+ /** Prepare without recording. Cleanup cancels a short press or hands off after start. */
+ prepare?: () => (() => void) | void;
/**
* 开始录音。返回 `false` 表示这次没有占用录音(已在录音、被取消等),
* 调用方不得再对当前会话执行 `stop()`。
@@ -68,11 +70,19 @@ export function createComposerLongPressVoiceGesture(
let stopInFlight = false;
let disposed = false;
let startGeneration = 0;
+ let releasePreparation: (() => void) | void;
+
+ const clearPreparation = (): void => {
+ const release = releasePreparation;
+ releasePreparation = undefined;
+ release?.();
+ };
const clearPending = (): void => {
if (!pending) return;
clearTimeout(pending.timer);
pending = null;
+ clearPreparation();
};
const endHold = (): void => {
@@ -109,14 +119,18 @@ export function createComposerLongPressVoiceGesture(
const beginVoiceInput = (): void => {
pending = null;
- if (disposed || !isIdleLike(options.getState())) return;
+ if (disposed || !isIdleLike(options.getState())) {
+ clearPreparation();
+ return;
+ }
holding = true;
options.onHoldStart?.();
const generation = ++startGeneration;
const currentStart = Promise.resolve()
.then(() => options.start())
.then((claimed) => claimed !== false)
- .catch(() => false);
+ .catch(() => false)
+ .finally(clearPreparation);
startPromise = currentStart;
void currentStart.then((claimed) => {
if (startPromise !== currentStart) return;
@@ -144,6 +158,11 @@ export function createComposerLongPressVoiceGesture(
origin: point,
timer: setTimeout(beginVoiceInput, holdMs),
};
+ try {
+ releasePreparation = options.prepare?.();
+ } catch {
+ // Preparation is optional; the normal start path can still recover.
+ }
},
move(point) {
if (!pending) return;
diff --git a/apps/desktop/src/renderer/voice-input/useVoiceInput.ts b/apps/desktop/src/renderer/voice-input/useVoiceInput.ts
index 68dfc6164e..fdad5e1576 100644
--- a/apps/desktop/src/renderer/voice-input/useVoiceInput.ts
+++ b/apps/desktop/src/renderer/voice-input/useVoiceInput.ts
@@ -44,6 +44,7 @@ import {
isSelectedMicrophoneUnavailableError,
prewarmVoiceInputBenchmarkFixture,
prewarmVoiceInputMicrophoneWithAutomaticFallback,
+ prepareVoiceInputMicrophone,
type PcmChunk,
} from './WebMicAudioEngine';
import { prewarmVoiceInputAudio } from './audioContextPool';
@@ -145,6 +146,7 @@ export type UseVoiceInputResult = {
getLastRefinement: () => VoiceInputRefinementSnapshot | null;
/** 同步读取当前状态,不经过 React 渲染。长按手势靠它判断是否占用了录音。 */
getState: () => VoiceInputState;
+ prepare: () => (() => void) | void;
/** 同步 claim 录音后立刻返回是否占用;启动流程(授权确认、采集、连接)在后台继续。 */
start: (options?: VoiceInputStartOptions) => boolean;
stop: (options?: VoiceInputStopOptions) => Promise<void>;
@@ -196,6 +198,7 @@ export function useVoiceInput(
const [state, setState] = useState<VoiceInputState>('idle');
const [captureReady, setCaptureReady] = useState(false);
const startupTimelineRef = useRef<VoiceInputStartupTimeline | null>(null);
+ const microphonePreparationRef = useRef<(() => void) | null>(null);
const [draftText, setDraftText] = useState('');
const [draftSource, setDraftSource] = useState<VoiceInputDraftSource | null>(null);
const [lastError, setLastError] = useState<string | null>(null);
@@ -1236,6 +1239,34 @@ export function useVoiceInput(
voiceInputSettings.microphoneDeviceId,
]);
+ const prepare = useCallback(() => {
+ if (disabledRef.current || !editorRef.current || editorRef.current.isDestroyed) return;
+ if (!['idle', 'done', 'error'].includes(stateRef.current)) return;
+ void prewarmVoiceInputAudio(workletUrl).catch(() => undefined);
+ void window.electronAPI.voiceInput.prewarm({
+ sourceLanguage: voiceInputSettings.language,
+ refinementEnabled: voiceInputSettings.refinementEnabled,
+ }).catch(() => undefined);
+ // Ordinary clicks must not trigger a new microphone permission prompt.
+ if (!window.electronAPI.voiceInput.getMicrophonePermissionCached().ok) return;
+ microphonePreparationRef.current?.();
+ const release = prepareVoiceInputMicrophone({
+ workletUrl,
+ deviceId: voiceInputSettings.microphoneDeviceId ?? undefined,
+ ...createVoiceInputAudioProfile(voiceInputSettings.fastActivationEnabled),
+ });
+ if (!release) return;
+ const cleanup = () => {
+ if (microphonePreparationRef.current === cleanup) microphonePreparationRef.current = null;
+ release();
+ };
+ microphonePreparationRef.current = cleanup;
+ return cleanup;
+ }, [voiceInputSettings.language, voiceInputSettings.refinementEnabled,
+ voiceInputSettings.microphoneDeviceId, voiceInputSettings.fastActivationEnabled]);
+
+ useEffect(() => () => { microphonePreparationRef.current?.(); }, []);
+
const start = useCallback((startOptions?: VoiceInputStartOptions): boolean => {
const currentState = stateRef.current;
if (
@@ -1280,7 +1311,8 @@ export function useVoiceInput(
label: '',
workletUrl,
deviceId: voiceInputSettings.microphoneDeviceId ?? undefined,
- fastActivationEnabled: voiceInputSettings.fastActivationEnabled,
+ fastActivationEnabled: voiceInputSettings.fastActivationEnabled || !!microphonePreparationRef.current,
+ temporaryMicrophone: !voiceInputSettings.fastActivationEnabled && !!microphonePreparationRef.current,
getRunId: () => runIdRef.current,
setEngine: (engine) => {
engineRef.current = engine;
@@ -1789,6 +1821,7 @@ export function useVoiceInput(
getLastSubmittedText: () => lastSubmittedTextRef.current,
getLastRefinement: () => lastRefinementRef.current,
getState,
+ prepare,
start,
stop: stopWithGate,
cancel,
diff --git a/apps/desktop/src/shared/voiceInputData.ts b/apps/desktop/src/shared/voiceInputData.ts
index f966f59afb..0c293e2646 100644
--- a/apps/desktop/src/shared/voiceInputData.ts
+++ b/apps/desktop/src/shared/voiceInputData.ts
@@ -178,10 +178,10 @@ export type VoiceInputDictionaryLearningEvidence = Pick<
export const DEFAULT_DICTIONARY_SYNC_ENABLED = true;
/**
- * 长按输入框语音输入的系统默认值。默认关闭:按住鼠标通常是在选字,只有用户主动
- * 打开才改变这个手势的含义。
+ * 长按输入框语音输入默认开启。设置页不展示开关;手动配置仍可通过
+ * composerLongPressEnabledOverride 显式关闭,删除 override 恢复默认。
*/
-export const DEFAULT_COMPOSER_LONG_PRESS_ENABLED = false;
+export const DEFAULT_COMPOSER_LONG_PRESS_ENABLED = true;
export const MAX_VOICE_INPUT_REFINEMENT_INSTRUCTIONS_CHARS = 1_000;
export const MAX_VOICE_INPUT_DICTIONARY_ENTRIES = 1_000;