[eric] voice: text-landed paste cue on successful injection, lock cue when a tap latches hands-free

frontend/src/shared/voice/VoiceDictationContext.tsx

@@ -1,6 +1,7 @@
 import React, { useCallback, useEffect, useRef } from 'react';
 import { useAppSelector } from '@/shared/hooks';
 import { useVoiceDictation } from './useVoiceDictation';
+import { playVoiceCue } from './voiceCues';
 import { VoiceContext } from './voiceContext';
 import VoiceOverlay from './VoiceOverlay';
 
@@ -29,13 +30,17 @@ export function VoiceDictationProvider({ children }: { children: React.ReactNode
   stateRef.current = state;
   const heldRef = useRef(false);
 
+  // A quick tap releases before start() flips state to 'recording': the session latches hands-free
+  // (Wispr's lock). The lock cue lands once recording is actually live, confirming the latch.
+  const latchedRef = useRef(false);
+
   const pressStart = useCallback((): void => {
     if (holdMode) {
       // A TAP while recording must stop: the quick tap's press-end fires before the async start
       // flips state to 'recording', so without this the mic could be started by a click but never
       // stopped by one.
       if (stateRef.current === 'recording') { heldRef.current = false; void stop(); return; }
-      if (stateRef.current === 'idle') { heldRef.current = true; void start(true); }
+      if (stateRef.current === 'idle') { heldRef.current = true; latchedRef.current = false; void start(true); }
     } else {
       toggle();
     }
@@ -45,9 +50,20 @@ export function VoiceDictationProvider({ children }: { children: React.ReactNode
     if (holdMode && heldRef.current) {
       heldRef.current = false;
       if (stateRef.current === 'recording') void stop();
+      else latchedRef.current = true;
     }
   }, [holdMode, stop]);
 
+  useEffect(() => {
+    if (state === 'recording' && latchedRef.current) {
+      latchedRef.current = false;
+      const t = setTimeout(() => playVoiceCue('lock'), 160);
+      return () => clearTimeout(t);
+    }
+    if (state === 'idle') latchedRef.current = false;
+    return undefined;
+  }, [state]);
+
   // Keyboard hotkey channels, matched to what the source can actually see:
   // voice:hold-down/up come ONLY from main's native uiohook tap (real global key-up, so the keyboard
   // gets the same hold-vs-toggle press semantics as the mic buttons); voice:toggle comes from the

frontend/src/shared/voice/useVoiceDictation.ts

@@ -223,9 +223,12 @@ export function useVoiceDictation() {
         // 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) {
+        if (target) {
+          playVoiceCue('paste');
+        } else {
           const inj = await window.openswarm?.voiceInject?.(text);
-          if (!inj?.pasted) setFeedback({ tone: 'ok', icon: 'clipboard', text: `${text}  (copied, press Cmd+V)`, at: Date.now() });
+          if (inj?.pasted) playVoiceCue('paste');
+          else setFeedback({ tone: 'ok', icon: 'clipboard', text: `${text}  (copied, press Cmd+V)`, at: Date.now() });
         }
         setState('idle');
       } else if (res?.ok && !res.text) {

frontend/src/shared/voice/voiceCues.ts

@@ -1,24 +1,35 @@
-import { VOICE_CUE_START, VOICE_CUE_STOP } from './voiceCueSounds';
+import { VOICE_CUE_START, VOICE_CUE_STOP, VOICE_CUE_PASTE, VOICE_CUE_LOCK } from './voiceCueSounds';
 
 // Eric picked these from Google's Material product sound set after a bake-off against Wispr Flow's
 // real cues; the synth versions never survived an ear test. Files are embedded data URIs (see
-// voiceCueSounds.ts), pre-instantiated so playback is instant on the press.
+// voiceCueSounds.ts), pre-instantiated so playback is instant on the press. The grammar mirrors
+// Wispr's three-beat: tap in, tap out, and a rising completion the moment the text actually lands;
+// lock marks a hands-free latch.
 
 const CUE_VOLUME = 0.35;
 
-const p_players: Record<'start' | 'stop', HTMLAudioElement | null> = { start: null, stop: null };
+type CueKind = 'start' | 'stop' | 'paste' | 'lock';
 
-function player(kind: 'start' | 'stop'): HTMLAudioElement {
+const SOURCES: Record<CueKind, string> = {
+  start: VOICE_CUE_START,
+  stop: VOICE_CUE_STOP,
+  paste: VOICE_CUE_PASTE,
+  lock: VOICE_CUE_LOCK,
+};
+
+const p_players: Partial<Record<CueKind, HTMLAudioElement>> = {};
+
+function player(kind: CueKind): HTMLAudioElement {
   let a = p_players[kind];
   if (!a) {
-    a = new Audio(kind === 'start' ? VOICE_CUE_START : VOICE_CUE_STOP);
-    a.volume = CUE_VOLUME;
+    a = new Audio(SOURCES[kind]);
+    a.volume = kind === 'paste' ? CUE_VOLUME * 0.8 : CUE_VOLUME;
     p_players[kind] = a;
   }
   return a;
 }
 
-export function playVoiceCue(kind: 'start' | 'stop'): void {
+export function playVoiceCue(kind: CueKind): void {
   try {
     const a = player(kind);
     a.currentTime = 0;