From 50ea107d9d1ba3ef57720362ce86bdb97a021b6c Mon Sep 17 00:00:00 2001
From: Teknium <127238744+teknium1@users.noreply.github.com>
Date: Thu, 10 Sep 2026 09:25:05 -0700
Subject: [PATCH] fix(desktop): read an auto-TTS reply aloud once when the HUD
 is open

With `voice.auto_tts` on and HUD mode active, a reply could be spoken
twice: the HUD renderer and the hidden app window both claim
`speak:<messageId>` through `hermes:ambient:claim`, and main collapsed
the two claims with the same 1 s deduper it uses for the turn-end beep.
The app window under the HUD is throttled by Chromium, so its transcript
subscription fires well past 1 s, finds the key pruned, and is told it
owns the cue. Fixes #99717.

The unit was wrong, not the window: a spoken reply is minutes of audio
keyed by a durable message id. `createAmbientClaimArbiter` keeps the
tick-sized window for `sound:*` and holds `speak:*` claims for a
10-minute TTL. No HUD special-casing: the same race exists between any
two windows showing the same chat, and the arbiter fixes them all.

Diagnosis credit: #99810 (@liuhao1024) and #100289 (@shivamjg101),
whose TTL framing this follows.
---
 apps/desktop/electron/event-dedupe.test.ts | 19 ++++++++++++++++++-
 apps/desktop/electron/event-dedupe.ts      | 18 ++++++++++++++++++
 apps/desktop/electron/main.ts              |  9 +++++----
 3 files changed, 41 insertions(+), 5 deletions(-)

diff --git a/apps/desktop/electron/event-dedupe.test.ts b/apps/desktop/electron/event-dedupe.test.ts
index 3c0f905587..c27096796b 100644
--- a/apps/desktop/electron/event-dedupe.test.ts
+++ b/apps/desktop/electron/event-dedupe.test.ts
@@ -2,7 +2,7 @@ import assert from 'node:assert/strict'
 
 import { test } from 'vitest'
 
-import { createEventDeduper } from './event-dedupe'
+import { createAmbientClaimArbiter, createEventDeduper } from './event-dedupe'
 
 test('collapses the same key inside the window (two windows, one event)', () => {
   const isDup = createEventDeduper(1000)
@@ -35,3 +35,20 @@ test('prunes stale keys so the map cannot grow unbounded', () => {
     assert.equal(isDup(`turnDone:s${i}`, i * 2000), false)
   }
 })
+
+// #99717: the hidden app window under an open HUD claims the same reply late.
+test('a spoken reply stays claimed long after the tick window, a beep does not', () => {
+  const owns = createAmbientClaimArbiter(1000)
+
+  assert.equal(owns('speak:m1', 0), true, 'HUD renderer claims the reply')
+  assert.equal(owns('speak:m1', 5_000), false, 'app window claiming 5 s later stays quiet')
+  assert.equal(owns('sound:turnDone:s1', 0), true)
+  assert.equal(owns('sound:turnDone:s1', 5_000), true, 'beep keys still re-fire after the tick window')
+})
+
+test('a spoken reply can be claimed again once the speech TTL elapses', () => {
+  const owns = createAmbientClaimArbiter(1000, 60_000)
+
+  assert.equal(owns('speak:m1', 0), true)
+  assert.equal(owns('speak:m1', 60_000), true)
+})
diff --git a/apps/desktop/electron/event-dedupe.ts b/apps/desktop/electron/event-dedupe.ts
index a2d12a8f2c..c8f553ed02 100644
--- a/apps/desktop/electron/event-dedupe.ts
+++ b/apps/desktop/electron/event-dedupe.ts
@@ -30,3 +30,21 @@ export function createEventDeduper(intervalMs = DEDUPE_INTERVAL_MS) {
     return false
   }
 }
+
+// A `speak:<messageId>` cue is seconds of audio keyed by a durable backend
+// message id, not an instant beep: the peer's claim can arrive well past the
+// 1 s window (the app window hidden under an open HUD is throttled by Chromium
+// and its transcript subscription fires late), and the reply was read twice
+// (#99717). One reply is one claim for as long as a reply can plausibly play.
+export const SPEECH_CLAIM_TTL_MS = 10 * 60_000
+
+// Cross-window arbiter for every ambient cue: `speak:*` keys hold for the
+// speech TTL, everything else keeps the tick-sized window.
+export function createAmbientClaimArbiter(intervalMs = DEDUPE_INTERVAL_MS, speechTtlMs = SPEECH_CLAIM_TTL_MS) {
+  const cues = createEventDeduper(intervalMs)
+  const speech = createEventDeduper(speechTtlMs)
+
+  return function owns(key: string, now = Date.now()): boolean {
+    return !(key.startsWith('speak:') ? speech(key, now) : cues(key, now))
+  }
+}
diff --git a/apps/desktop/electron/main.ts b/apps/desktop/electron/main.ts
index 4fb88c2d4d..d0c2e18513 100644
--- a/apps/desktop/electron/main.ts
+++ b/apps/desktop/electron/main.ts
@@ -169,7 +169,7 @@ import {
 } from './desktop-uninstall'
 import { describeDevCdpDecision, resolveDevCdpPort } from './dev-cdp'
 import { installEmbedReferer } from './embed-referer'
-import { createEventDeduper } from './event-dedupe'
+import { createAmbientClaimArbiter } from './event-dedupe'
 import {
   buildTerminalScript,
   resolveTerminalLaunch,
@@ -16799,9 +16799,10 @@ ipcMain.handle('hermes:api', async (_event, request) => {
   return handleHermesApiRequest(request).finally(releaseProfileDeletion)
 })
 
-// Main serializes cross-window ambient claims.
-const claimedAmbientCue = createEventDeduper()
-ipcMain.handle('hermes:ambient:claim', (_event, key) => !claimedAmbientCue(String(key ?? '')))
+// Main serializes cross-window ambient claims (see event-dedupe.ts for why a
+// spoken reply holds its claim far longer than a beep).
+const ownsAmbientCue = createAmbientClaimArbiter()
+ipcMain.handle('hermes:ambient:claim', (_event, key) => ownsAmbientCue(String(key ?? '')))
 
 registerNativeNotifications({ getMainWindow: () => mainWindow, focusWindow })