[eric] voice: F5 mic key as the simple hotkey, remove the activation sounds

electron/main.js

@@ -1793,18 +1793,22 @@ app.whenReady().then(async () => {
   // focused, MAIN watches before-input-event (sees every keyDown/keyUp incl. modifiers, regardless of
   // which element or webview has focus, and immune to macOS's letter-keyup-under-Cmd suppression at the
   // DOM layer) and relays voice:hold-down / voice:hold-up; the renderer maps those to hold-vs-toggle.
-  const VOICE_COMBO = 'CommandOrControl+Shift+D';
+  // F5 is the mic/dictation key printed on Mac keyboards (the fn/globe key itself is invisible to
+  // Electron without a native event tap), so it's the simple primary; the old combo stays as backup.
+  const VOICE_COMBOS = ['F5', 'CommandOrControl+Shift+D'];
   const registerVoiceShortcut = () => {
-    try {
-      if (!globalShortcut.isRegistered(VOICE_COMBO)) {
-        globalShortcut.register(VOICE_COMBO, () => {
-          if (mainWindow && !mainWindow.isDestroyed()) mainWindow.webContents.send('voice:toggle');
-        });
-      }
-    } catch (_) { /* a taken shortcut just means no global hotkey; the pill still works */ }
+    for (const combo of VOICE_COMBOS) {
+      try {
+        if (!globalShortcut.isRegistered(combo)) {
+          globalShortcut.register(combo, () => {
+            if (mainWindow && !mainWindow.isDestroyed()) mainWindow.webContents.send('voice:toggle');
+          });
+        }
+      } catch (_) { /* a taken shortcut just means no global hotkey; the pill still works */ }
+    }
   };
   registerVoiceShortcut();
-  app.on('browser-window-focus', () => { try { globalShortcut.unregister(VOICE_COMBO); } catch (_) {} });
+  app.on('browser-window-focus', () => { for (const combo of VOICE_COMBOS) { try { globalShortcut.unregister(combo); } catch (_) {} } });
   app.on('browser-window-blur', registerVoiceShortcut);
 
   // Every discrete combo press relays (autorepeat filtered); the renderer toggles record state.
@@ -1813,8 +1817,10 @@ app.whenReady().then(async () => {
   // without a native event tap. Keyboard = press to start/stop; the mic buttons own true hold-to-talk.
   const installVoiceHoldRelay = (contents) => {
     contents.on('before-input-event', (event, input) => {
+      if (input.type !== 'keyDown' || input.isAutoRepeat) return;
       const isD = (input.code === 'KeyD' || (input.key || '').toLowerCase() === 'd');
-      if (input.type === 'keyDown' && isD && (input.meta || input.control) && input.shift && !input.isAutoRepeat) {
+      const combo = (isD && (input.meta || input.control) && input.shift) || input.code === 'F5';
+      if (combo) {
         if (mainWindow && !mainWindow.isDestroyed()) mainWindow.webContents.send('voice:hold-down');
         event.preventDefault();
       }

frontend/src/app/pages/Dashboard/desktop/DesktopSpawnPill.tsx

@@ -160,7 +160,7 @@ function DesktopSpawnPill({
         >
           <AddRounded sx={{ fontSize: 18 }} />
         </Box>
-        <Tooltip title={recording ? 'Stop dictation' : preparing ? `Downloading voice model ${voicePct}%` : 'Dictate (Cmd+Shift+D)'} placement="top" arrow>
+        <Tooltip title={recording ? 'Stop dictation' : preparing ? `Downloading voice model ${voicePct}%` : 'Dictate (F5)'} placement="top" arrow>
           <Box
             role="button"
             aria-label="Voice dictation"

frontend/src/app/pages/Settings/sections/general/GeneralInterface.tsx

@@ -104,7 +104,7 @@ const GeneralInterface: React.FC<{
       <Box sx={inlineRowSx} {...settingSelectAttrs('voice_hold_to_talk', 'Dictation', 'Interface', 'Hold to talk, or tap to start and stop.')}>
         <Box sx={{ mr: 3 }}>
           <Typography sx={labelSx}>Dictation</Typography>
-          <Typography sx={descSx}>How the mic button works. Cmd+Shift+D always starts and stops.</Typography>
+          <Typography sx={descSx}>How the mic button works. The mic key (F5) starts and stops.</Typography>
         </Box>
         <ToggleButtonGroup
           value={form.voice_hold_to_talk ?? true}

frontend/src/shared/voice/useVoiceDictation.ts

@@ -2,7 +2,6 @@ import { useCallback, useEffect, useRef, useState } from 'react';
 import { API_BASE } from '@/shared/config';
 import { encodeWav, VOICE_SAMPLE_RATE } from './encodeWav';
 import { injectAtFocus } from './injectAtFocus';
-import { playCancel, playDone, playStart } from './voiceSounds';
 
 export type VoiceState = 'idle' | 'recording' | 'transcribing' | 'preparing';
 
@@ -118,7 +117,6 @@ export function useVoiceDictation() {
       node.connect(ctx.destination);
       recRef.current = { ctx, stream, node, source, chunks };
       setState('recording');
-      playStart();
       // Warm the model the moment recording begins so transcription is instant on stop.
       void window.openswarm?.voiceWarmup?.();
     } catch (err) {
@@ -145,19 +143,15 @@ export function useVoiceDictation() {
         setLastText(text);
         // Land the text where the user's cursor is: focused field, then focused browser page, then
         // the OS paste fallback (other apps). The floating bubble is just confirmation, not the output.
-        // Success is silent-visual: the text landing at the cursor IS the feedback (plus the done blip).
-        // Only the clipboard fallback still speaks, because the user has to act (paste) to get the text.
+        // Success is silent: the text landing at the cursor IS the feedback. Only the clipboard
+        // fallback still speaks, because the user has to act (paste) to get the text.
         const target = injectAtFocus(text);
-        if (target) {
-          playDone();
-        } else {
+        if (!target) {
           const inj = await window.openswarm?.voiceInject?.(text);
-          if (inj?.pasted) playDone();
-          else setFeedback({ tone: 'ok', icon: 'clipboard', text: `${text}  (copied, press Cmd+V)`, at: Date.now() });
+          if (!inj?.pasted) setFeedback({ tone: 'ok', icon: 'clipboard', text: `${text}  (copied, press Cmd+V)`, at: Date.now() });
         }
         setState('idle');
       } else if (res?.ok && !res.text) {
-        playCancel();
         setFeedback({ tone: 'warn', icon: 'info', text: "Didn't catch that. Try again.", at: Date.now() });
         setState('idle');
       } else if (res?.error === 'model-downloading' || res?.error === 'no-model') {

frontend/src/shared/voice/voiceSounds.ts

@@ -1,35 +0,0 @@
-// Tiny WebAudio cues for dictation, generated in code (no bundled assets): a soft rising two-note on
-// start, falling on cancel, and a quiet tick when text lands. Volumes stay whisper-level on purpose.
-let ctx: AudioContext | null = null;
-
-function audio(): AudioContext | null {
-  try {
-    if (!ctx) ctx = new AudioContext();
-    if (ctx.state === 'suspended') void ctx.resume();
-    return ctx;
-  } catch {
-    return null;
-  }
-}
-
-function blip(freqs: Array<[number, number, number]>): void {
-  const ac = audio();
-  if (!ac) return;
-  const t0 = ac.currentTime;
-  for (const [freq, at, dur] of freqs) {
-    const osc = ac.createOscillator();
-    const gain = ac.createGain();
-    osc.type = 'sine';
-    osc.frequency.value = freq;
-    gain.gain.setValueAtTime(0, t0 + at);
-    gain.gain.linearRampToValueAtTime(0.055, t0 + at + 0.02);
-    gain.gain.exponentialRampToValueAtTime(0.0001, t0 + at + dur);
-    osc.connect(gain).connect(ac.destination);
-    osc.start(t0 + at);
-    osc.stop(t0 + at + dur + 0.05);
-  }
-}
-
-export function playStart(): void { blip([[523, 0, 0.12], [784, 0.09, 0.16]]); }
-export function playCancel(): void { blip([[784, 0, 0.1], [523, 0.08, 0.14]]); }
-export function playDone(): void { blip([[1047, 0, 0.09]]); }