fix(android): stop push-to-talk from dropping speech after mid-hold pauses (#110995)

* fix(android): stop push-to-talk from dropping speech after mid-hold pauses The SpeechRecognizer PTT path ran one session with a 2.5s silence window, kept only the last final, and never restarted during a hold - speech after a mid-hold pause was silently dropped. Capture now uses an ordered rung ladder: API 33+ app-owned AudioRecord piped via EXTRA_AUDIO_SOURCE with EXTRA_SEGMENTED_SESSION, then silence-keyed segmented sessions, then a restart-during-hold single-session fallback for older devices; failed holds only descend. Final segments accumulate and merge locale-aware (no ASCII separators after CJK/Arabic punctuation), with a trailing live partial deduped against the last final. Release waits one bounded grace for the terminal recognizer callback so late finals are not truncated; rapid re-press drains the prior release instead of cancelling it; cancellation mid-release still tears capture down, resumes realtime talk, and resolves the stop payload. Raw-rung release also stops the recognizer since EXTRA_AUDIO_SOURCE is optional and a service may run its own microphone. Input level meters from our PCM stream in the raw rung where onRmsChanged never fires. * fix(android): API-33 recognizer extras behind RequiresApi and inventory refresh Min-SDK lint (InlinedApi) flagged the API 33 RecognizerIntent extras; they now live in RequiresApi(TIRAMISU) helpers behind explicit SDK gates, and the native i18n inventory is refreshed for the moved lines. * chore(android): rebuild native locale artifacts for inventory order
2026-08-12 21:53:00 -06:00 · 2026-07-19 00:13:19 -07:00
parent ea54060223
commit c84a59284b
28 changed files with 1045 additions and 194 deletions
@@ -13043,7 +13043,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 532,
+      "line": 602,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Speech recognizer unavailable",
      "surface": "android",
@@ -13051,15 +13051,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 596,
-      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
-      "source": "Listening (PTT)",
-      "surface": "android",
-      "id": "native.android.0faf9df3e5f708df"
-    },
-    {
-      "kind": "ui-call",
-      "line": 693,
+      "line": 800,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Ready",
      "surface": "android",
@@ -13067,7 +13059,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 911,
+      "line": 1018,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Start failed: $message",
      "surface": "android",
@@ -13075,7 +13067,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 1001,
+      "line": 1114,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Connecting…",
      "surface": "android",
@@ -13083,7 +13075,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 1070,
+      "line": 1183,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Off",
      "surface": "android",
@@ -13091,7 +13083,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 1073,
+      "line": 1186,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Talk failed: Realtime provider closed unexpectedly.",
      "surface": "android",
@@ -13099,7 +13091,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 1078,
+      "line": 1191,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Talk failed: Realtime provider closed: $reason",
      "surface": "android",
@@ -13107,7 +13099,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 2216,
+      "line": 2546,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Thinking…",
      "surface": "android",
@@ -13115,7 +13107,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 2233,
+      "line": 2563,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Gateway not connected",
      "surface": "android",
@@ -13123,7 +13115,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 2246,
+      "line": 2576,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Aborted",
      "surface": "android",
@@ -13131,7 +13123,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 2246,
+      "line": 2576,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Chat error",
      "surface": "android",
@@ -13139,7 +13131,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 2263,
+      "line": 2593,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "No reply",
      "surface": "android",
@@ -13147,7 +13139,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 2279,
+      "line": 2609,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Talk failed: $message",
      "surface": "android",
@@ -13155,7 +13147,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 2504,
+      "line": 2902,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Generating voice…",
      "surface": "android",
@@ -13163,7 +13155,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 2532,
+      "line": 2930,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Speak failed: $message",
      "surface": "android",
@@ -13171,7 +13163,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 2681,
+      "line": 3079,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Speaking…",
      "surface": "android",
@@ -13179,7 +13171,15 @@
    },
    {
      "kind": "ui-call",
-      "line": 2998,
+      "line": 3430,
+      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
+      "source": "Listening (PTT)",
+      "surface": "android",
+      "id": "native.android.0faf9df3e5f708df"
+    },
+    {
+      "kind": "ui-call",
+      "line": 3441,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Microphone permission required",
      "surface": "android",
@@ -13187,7 +13187,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 3004,
+      "line": 3447,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Audio error",
      "surface": "android",
@@ -13195,7 +13195,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 3005,
+      "line": 3448,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Client error",
      "surface": "android",
@@ -13203,7 +13203,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 3006,
+      "line": 3449,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Network error",
      "surface": "android",
@@ -13211,7 +13211,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 3007,
+      "line": 3450,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Network timeout",
      "surface": "android",
@@ -13219,7 +13219,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 3009,
+      "line": 3452,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Recognizer busy",
      "surface": "android",
@@ -13227,7 +13227,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 3010,
+      "line": 3453,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Server error",
      "surface": "android",
@@ -13235,7 +13235,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 3011,
+      "line": 3454,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Listening",
      "surface": "android",
@@ -13243,7 +13243,7 @@
    },
    {
      "kind": "ui-call",
-      "line": 3012,
+      "line": 3455,
      "path": "apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt",
      "source": "Speech error ($error)",
      "surface": "android",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "أداة التعرّف على الكلام غير متاحة"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "جارٍ الاستماع (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "جارٍ التحدث…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "جارٍ الاستماع (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "Spracherkennung nicht verfügbar"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "Hört zu (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "Spricht…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "Hört zu (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "El reconocimiento de voz no está disponible"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "Escuchando (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "Hablando…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "Escuchando (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "تشخیص‌دهنده گفتار در دسترس نیست"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "در حال شنیدن (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "در حال صحبت…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "در حال شنیدن (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "Reconnaissance vocale indisponible"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "Écoute (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "Lecture en cours…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "Écoute (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "वाक् पहचानकर्ता उपलब्ध नहीं है"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "सुन रहा है (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "बोल रहा है…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "सुन रहा है (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "Pengenal ucapan tidak tersedia"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "Mendengarkan (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "Berbicara…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "Mendengarkan (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "Riconoscimento vocale non disponibile"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "In ascolto (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "Riproduzione vocale…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "In ascolto (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "音声認識を利用できません"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "聞き取り中（PTT）"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "読み上げ中…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "聞き取り中（PTT）"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "음성 인식기를 사용할 수 없습니다"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "듣는 중(PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "말하는 중…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "듣는 중(PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "Spraakherkenning niet beschikbaar"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "Luisteren (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "Spreken…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "Luisteren (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "Rozpoznawanie mowy jest niedostępne"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "Nasłuchiwanie (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "Mówienie…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "Nasłuchiwanie (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "Reconhecimento de fala indisponível"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "Ouvindo (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "Falando…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "Ouvindo (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "Распознавание речи недоступно"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "Прослушивание (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "Воспроизведение речи…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "Прослушивание (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "Taligenkänning är inte tillgänglig"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "Lyssnar (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "Talar…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "Lyssnar (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "ไม่สามารถใช้ตัวจดจำเสียงพูดได้"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "กำลังฟัง (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "กำลังพูด…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "กำลังฟัง (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "Konuşma tanıyıcı kullanılamıyor"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "Dinleniyor (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "Konuşuyor…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "Dinleniyor (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "Розпізнавання мовлення недоступне"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "Прослуховування (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "Озвучення…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "Прослуховування (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "Trình nhận dạng giọng nói không khả dụng"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "Đang nghe (PTT)"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "Đang nói…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "Đang nghe (PTT)"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "语音识别器不可用"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "正在聆听（按住说话）"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "正在朗读…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "正在聆听（按住说话）"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -8158,11 +8158,6 @@
      "source": "Speech recognizer unavailable",
      "translated": "語音辨識器無法使用"
    },
-    {
-      "id": "native.android.0faf9df3e5f708df",
-      "source": "Listening (PTT)",
-      "translated": "正在聆聽（PTT）"
-    },
    {
      "id": "native.android.47a59f8798682ea0",
      "source": "Ready",
@@ -8238,6 +8233,11 @@
      "source": "Speaking…",
      "translated": "正在朗讀…"
    },
+    {
+      "id": "native.android.0faf9df3e5f708df",
+      "source": "Listening (PTT)",
+      "translated": "正在聆聽（PTT）"
+    },
    {
      "id": "native.android.d1a7db2c16fd6d09",
      "source": "Microphone permission required",
@@ -0,0 +1,25 @@
+package ai.openclaw.app.voice
+
+internal enum class PushToTalkRecognitionCandidate {
+  RawAudioSegmented,
+  SilenceSegmented,
+  RestartingSingleSession,
+}
+
+internal fun pushToTalkRecognitionCandidates(
+  supportsSegmentedRecognition: Boolean,
+  first: PushToTalkRecognitionCandidate?,
+): List<PushToTalkRecognitionCandidate> {
+  val available =
+    if (supportsSegmentedRecognition) {
+      PushToTalkRecognitionCandidate.entries
+    } else {
+      listOf(PushToTalkRecognitionCandidate.RestartingSingleSession)
+    }
+  if (first == null) return available
+  return available.dropWhile { it != first }.ifEmpty { listOf(PushToTalkRecognitionCandidate.RestartingSingleSession) }
+}
+
+internal fun shouldAdvancePushToTalkRungAfterSegmentedSession(
+  candidate: PushToTalkRecognitionCandidate,
+): Boolean = candidate == PushToTalkRecognitionCandidate.RawAudioSegmented
@@ -0,0 +1,39 @@
+package ai.openclaw.app.voice
+
+import java.util.Locale
+
+internal object PushToTalkTranscriptMerger {
+  private val trailingClosers = setOf('"', '\'', '’', '”', ')', ']', '}')
+
+  fun merge(
+    finalSegments: List<String>,
+    livePartial: String?,
+  ): String {
+    val segments = finalSegments.map { it.trim() }.filter { it.isNotEmpty() }.toMutableList()
+    val partial = livePartial?.trim().orEmpty()
+    val lastFinal = segments.lastOrNull()
+    if (partial.isNotEmpty() && (lastFinal == null || normalize(partial) != normalize(lastFinal))) {
+      segments += partial
+    }
+
+    return buildString {
+      var previousEndsSentence = false
+      segments.forEachIndexed { index, segment ->
+        if (index > 0) {
+          append(if (previousEndsSentence) " " else ". ")
+        }
+        append(segment)
+        // Recognizers emit locale punctuation (。？؟…); injecting ASCII ". " after those corrupts
+        // the transcript. Separate with ". " only when a segment ends in a letter or digit.
+        val finalCharacter = segment.trimEnd { it in trailingClosers }.lastOrNull()
+        previousEndsSentence = finalCharacter != null && !finalCharacter.isLetterOrDigit()
+      }
+    }
+  }
+
+  private fun normalize(value: String): String =
+    value
+      .trim()
+      .lowercase(Locale.ROOT)
+      .replace(Regex("""\s+"""), " ")
+}
@@ -17,10 +17,14 @@ import android.media.AudioAttributes
 import android.media.AudioFocusRequest
 import android.media.AudioFormat
 import android.media.AudioManager
+import android.media.AudioRecord
 import android.media.AudioTrack
+import android.media.MediaRecorder
+import android.os.Build
 import android.os.Bundle
 import android.os.Handler
 import android.os.Looper
+import android.os.ParcelFileDescriptor
 import android.os.SystemClock
 import android.speech.RecognitionListener
 import android.speech.RecognizerIntent
@@ -29,6 +33,7 @@ import android.speech.tts.TextToSpeech
 import android.speech.tts.UtteranceProgressListener
 import android.util.Base64
 import android.util.Log
+import androidx.annotation.RequiresApi
 import androidx.core.content.ContextCompat
 import kotlinx.coroutines.CancellationException
 import kotlinx.coroutines.CompletableDeferred
@@ -53,15 +58,19 @@ import kotlinx.coroutines.sync.Mutex
 import kotlinx.coroutines.sync.withLock
 import kotlinx.coroutines.withContext
 import kotlinx.coroutines.withTimeout
+import kotlinx.coroutines.withTimeoutOrNull
+import kotlinx.coroutines.yield
 import kotlinx.serialization.json.Json
 import kotlinx.serialization.json.JsonArray
 import kotlinx.serialization.json.JsonElement
 import kotlinx.serialization.json.JsonObject
 import kotlinx.serialization.json.JsonPrimitive
 import kotlinx.serialization.json.buildJsonObject
+import java.io.IOException
 import java.util.LinkedHashMap
 import java.util.Locale
 import java.util.UUID
+import java.util.concurrent.atomic.AtomicBoolean
 import java.util.concurrent.atomic.AtomicLong
 import kotlin.coroutines.coroutineContext

@@ -183,6 +192,57 @@ private data class PendingRealtimePlaybackMark(
  var targetFrame: Long? = null,
 )

+private class PushToTalkAudioSource(
+  val readDescriptor: ParcelFileDescriptor,
+  private val writeStream: ParcelFileDescriptor.AutoCloseOutputStream,
+  private val audioRecord: AudioRecord,
+) {
+  private val finishRequested = AtomicBoolean(false)
+  private val inputFinished = AtomicBoolean(false)
+  private val descriptorClosed = AtomicBoolean(false)
+  var pumpJob: Job? = null
+
+  fun requestFinish() {
+    if (!finishRequested.compareAndSet(false, true)) return
+    runCatching { audioRecord.stop() }
+  }
+
+  fun finishFromPump() {
+    if (!inputFinished.compareAndSet(false, true)) return
+    runCatching { audioRecord.stop() }
+    runCatching { audioRecord.release() }
+    runCatching { writeStream.close() }
+  }
+
+  fun close() {
+    requestFinish()
+    pumpJob?.cancel()
+    finishFromPump()
+    pumpJob = null
+    if (descriptorClosed.compareAndSet(false, true)) {
+      runCatching { readDescriptor.close() }
+    }
+  }
+}
+
+private sealed interface PushToTalkRecognitionRung {
+  val candidate: PushToTalkRecognitionCandidate
+
+  data class RawAudioSegmented(
+    val source: PushToTalkAudioSource,
+  ) : PushToTalkRecognitionRung {
+    override val candidate = PushToTalkRecognitionCandidate.RawAudioSegmented
+  }
+
+  data object SilenceSegmented : PushToTalkRecognitionRung {
+    override val candidate = PushToTalkRecognitionCandidate.SilenceSegmented
+  }
+
+  data object RestartingSingleSession : PushToTalkRecognitionRung {
+    override val candidate = PushToTalkRecognitionCandidate.RestartingSingleSession
+  }
+}
+
 class TalkModeManager internal constructor(
  private val context: Context,
  private val scope: CoroutineScope,
@@ -209,6 +269,10 @@ class TalkModeManager internal constructor(
    private const val maxConversationEntries = 40
    private const val realtimePlaybackBufferMs = 240
    private const val realtimeUserFinalRewriteGraceMs = 1_500L
+    private const val pushToTalkSampleRateHz = 16_000
+    private const val pushToTalkReleaseGraceMs = 5_000L
+    private const val pushToTalkReleaseDrainTimeoutMs = 6_000L
+    private const val pushToTalkRestartDelayMs = 200L
  }

  private val mainHandler = Handler(Looper.getMainLooper())
@@ -288,6 +352,10 @@ class TalkModeManager internal constructor(
  private var pttAutoStopEnabled = false
  private var pttTimeoutJob: Job? = null
  private var pttCompletion: CompletableDeferred<TalkPttStopPayload>? = null
+  private var pttRecognitionRung: PushToTalkRecognitionRung? = null
+  private var pttReleaseCompletion: CompletableDeferred<Unit>? = null
+  private val pttFinalSegments = mutableListOf<String>()
+  private var pttLivePartial = ""

  private var silenceJob: Job? = null
  private var silenceWindowMs = TalkDefaults.defaultSilenceTimeoutMs
@@ -511,7 +579,9 @@ class TalkModeManager internal constructor(
    }
    // PTT begin is idempotent so gateway retries don't start multiple recognizers.
    activePttCaptureId?.let {
-      return PushToTalkStartResult.Existing(TalkPttStartPayload(captureId = it))
+      if (pttReleaseCompletion == null) {
+        return PushToTalkStartResult.Existing(TalkPttStartPayload(captureId = it))
+      }
    }
    finishingPttCaptureId?.let {
      throw IllegalStateException("PTT_BUSY: previous push-to-talk turn is still finishing")
@@ -537,8 +607,14 @@ class TalkModeManager internal constructor(
    val captureGeneration = startGeneration.get()
    return try {
      withContext(Dispatchers.Main) {
+        val hasPendingRelease = pttReleaseCompletion != null
+        if (hasPendingRelease) {
+          drainPushToTalkReleaseBeforeBegin()
+        }
        activePttCaptureId?.let {
-          return@withContext PushToTalkStartResult.Existing(TalkPttStartPayload(captureId = it))
+          if (!hasPendingRelease) {
+            return@withContext PushToTalkStartResult.Existing(TalkPttStartPayload(captureId = it))
+          }
        }
        finishingPttCaptureId?.let {
          throw IllegalStateException("PTT_BUSY: previous push-to-talk turn is still finishing")
@@ -559,6 +635,10 @@ class TalkModeManager internal constructor(
        recognizer?.cancel()
        recognizer?.destroy()
        recognizer = null
+        closePushToTalkRung()
+        pttReleaseCompletion = null
+        pttFinalSegments.clear()
+        pttLivePartial = ""
        lastTranscript = ""
        lastHeardAtMs = null
        activePttCaptureId = captureId
@@ -577,9 +657,13 @@ class TalkModeManager internal constructor(
          ) {
            throw IllegalStateException("NODE_BACKGROUND_UNAVAILABLE: command requires foreground")
          }
-          recognizer = SpeechRecognizer.createSpeechRecognizer(context).also { it.setRecognitionListener(listener) }
-          startListeningInternal(markListening = true)
+          recognizer =
+            SpeechRecognizer.createSpeechRecognizer(context).also {
+              it.setRecognitionListener(recognitionListener(captureId))
+            }
+          startPushToTalkRecognition(captureId)
        } catch (err: Throwable) {
+          closePushToTalkRung()
          runCatching { recognizer?.cancel() }
          runCatching { recognizer?.destroy() }
          recognizer = null
@@ -624,59 +708,82 @@ class TalkModeManager internal constructor(
  }

  internal suspend fun endPushToTalk(captureId: String): TalkPttStopPayload =
-    withContext(Dispatchers.Main) {
-      val cleared =
-        clearPushToTalkRecognition(captureId)
-          ?: return@withContext TalkPttStopPayload(captureId = captureId, transcript = null, status = "idle")
-      val transcript = cleared.transcript
+    try {
+      withContext(Dispatchers.Main) {
+        awaitPushToTalkRelease(captureId)
+        val cleared =
+          clearPushToTalkRecognition(captureId)
+            ?: return@withContext TalkPttStopPayload(captureId = captureId, transcript = null, status = "idle")
+        val transcript = cleared.transcript

-      if (transcript.isEmpty()) {
-        setStatus(if (_isEnabled.value) nativeText("Listening") else nativeText("Ready"))
-        resumeRealtimeCaptureAfterPushToTalk(captureId)
-        return@withContext finishPushToTalk(
-          TalkPttStopPayload(captureId = captureId, transcript = null, status = "empty"),
-          cleared.completion,
-        )
-      }
+        if (transcript.isEmpty()) {
+          setStatus(if (_isEnabled.value) nativeText("Listening") else nativeText("Ready"))
+          resumeRealtimeCaptureAfterPushToTalk(captureId)
+          return@withContext finishPushToTalk(
+            TalkPttStopPayload(captureId = captureId, transcript = null, status = "empty"),
+            cleared.completion,
+          )
+        }

-      if (!isConnected()) {
-        setStatus(nativeText("Gateway not connected"))
-        resumeRealtimeCaptureAfterPushToTalk(captureId)
-        return@withContext finishPushToTalk(
-          TalkPttStopPayload(captureId = captureId, transcript = transcript, status = "offline"),
-          cleared.completion,
-        )
-      }
+        if (!isConnected()) {
+          setStatus(nativeText("Gateway not connected"))
+          resumeRealtimeCaptureAfterPushToTalk(captureId)
+          return@withContext finishPushToTalk(
+            TalkPttStopPayload(captureId = captureId, transcript = transcript, status = "offline"),
+            cleared.completion,
+          )
+        }

-      setStatus(nativeText("Thinking…"), awaitingAgent = true)
-      lateinit var finishingJob: Job
-      finishingJob =
-        // Gateway-scoped so a switch drops the stale finalize; the NonCancellable
-        // finally still resumes capture when the scope cancels this job.
-        gatewayWorkScope.launch(start = CoroutineStart.LAZY) {
-          try {
-            finalizeTranscript(transcript)
-          } finally {
-            withContext(NonCancellable + Dispatchers.Main) {
-              resumeRealtimeCaptureAfterPushToTalk(captureId)
-              clearFinishingPushToTalk(captureId, finishingJob)
+        setStatus(nativeText("Thinking…"), awaitingAgent = true)
+        lateinit var finishingJob: Job
+        finishingJob =
+          // Gateway-scoped so a switch drops the stale finalize; the NonCancellable
+          // finally still resumes capture when the scope cancels this job.
+          gatewayWorkScope.launch(start = CoroutineStart.LAZY) {
+            try {
+              finalizeTranscript(transcript)
+            } finally {
+              withContext(NonCancellable + Dispatchers.Main) {
+                resumeRealtimeCaptureAfterPushToTalk(captureId)
+                clearFinishingPushToTalk(captureId, finishingJob)
+              }
            }
          }
+        // Cancellation can win before a lazy coroutine enters its body, in which
+        // case its finally block never runs. Completion still releases ownership.
+        finishingJob.invokeOnCompletion { clearFinishingPushToTalk(captureId, finishingJob) }
+        // Publish the job before it can run so stop() cannot clear ownership while
+        // an untracked finalizer still uses shared chat and playback state.
+        synchronized(finishingPttLock) {
+          finishingPttCaptureId = captureId
+          finishingPttJob = finishingJob
+          finishingJob.start()
        }
-      // Cancellation can win before a lazy coroutine enters its body, in which
-      // case its finally block never runs. Completion still releases ownership.
-      finishingJob.invokeOnCompletion { clearFinishingPushToTalk(captureId, finishingJob) }
-      // Publish the job before it can run so stop() cannot clear ownership while
-      // an untracked finalizer still uses shared chat and playback state.
-      synchronized(finishingPttLock) {
-        finishingPttCaptureId = captureId
-        finishingPttJob = finishingJob
-        finishingJob.start()
+        finishPushToTalk(
+          TalkPttStopPayload(captureId = captureId, transcript = transcript, status = "queued"),
+          cleared.completion,
+        )
      }
-      finishPushToTalk(
-        TalkPttStopPayload(captureId = captureId, transcript = transcript, status = "queued"),
-        cleared.completion,
-      )
+    } catch (err: CancellationException) {
+      // Mirror the normal termination tail: resume realtime capture, restore status, and
+      // resolve the PTT completion so a cancelled end (gateway drop) cannot leave Talk
+      // paused or an awaiter stuck behind the release wait.
+      withContext(NonCancellable + Dispatchers.Main) {
+        val cleared = clearPushToTalkRecognition(captureId)
+        if (cleared != null) {
+          setStatus(if (_isEnabled.value) nativeText("Listening") else nativeText("Ready"))
+          resumeRealtimeCaptureAfterPushToTalk(captureId)
+          finishPushToTalk(
+            TalkPttStopPayload(
+              captureId = captureId,
+              transcript = cleared.transcript.ifEmpty { null },
+              status = "cancelled",
+            ),
+            cleared.completion,
+          )
+        }
+      }
+      throw err
    }

  /** Cancels push-to-talk capture without sending the current transcript. */
@@ -934,9 +1041,15 @@ class TalkModeManager internal constructor(
    restartJob = null
    silenceJob?.cancel()
    silenceJob = null
+    closePushToTalkRung()
+    pttReleaseCompletion?.cancel()
+    pttReleaseCompletion = null
+    pttFinalSegments.clear()
+    pttLivePartial = ""
    lastTranscript = ""
    lastHeardAtMs = null
    _isListening.value = false
+    _inputLevel.value = 0f
    setStatus(nativeText("Off"), state = TalkStatusState.Off)
    stopRealtimeRelay()
    stopSpeaking()
@@ -2105,6 +2218,205 @@ class TalkModeManager internal constructor(
  private val transcriptSpaceAfterPunctuation =
    setOf('.', '!', '?', ',', ':', ';', ')', ']', '}', '"', '\'', '’', '”')

+  // API 33 adds segmented callbacks and caller-owned audio. Keep this ordered ladder
+  // in one place: removing the restart rung makes older devices drop speech after a pause.
+  private fun pushToTalkCandidates(first: PushToTalkRecognitionCandidate?): List<PushToTalkRecognitionCandidate> =
+    pushToTalkRecognitionCandidates(
+      supportsSegmentedRecognition = Build.VERSION.SDK_INT >= Build.VERSION_CODES.TIRAMISU,
+      first = first,
+    )
+
+  private fun startPushToTalkRecognition(
+    captureId: String,
+    firstCandidate: PushToTalkRecognitionCandidate? = null,
+  ) {
+    val recognizerInstance = recognizer ?: error("Speech recognizer unavailable")
+    var lastFailure: Throwable? = null
+    for (candidate in pushToTalkCandidates(firstCandidate)) {
+      try {
+        val rung =
+          when (candidate) {
+            PushToTalkRecognitionCandidate.RawAudioSegmented ->
+              PushToTalkRecognitionRung.RawAudioSegmented(openPushToTalkAudioSource())
+            PushToTalkRecognitionCandidate.SilenceSegmented -> PushToTalkRecognitionRung.SilenceSegmented
+            PushToTalkRecognitionCandidate.RestartingSingleSession ->
+              PushToTalkRecognitionRung.RestartingSingleSession
+          }
+        pttRecognitionRung = rung
+        recognizerInstance.startListening(pushToTalkRecognizerIntent(rung))
+        _isListening.value = true
+        setStatus(nativeText("Listening (PTT)"))
+        return
+      } catch (err: Throwable) {
+        lastFailure = err
+        closePushToTalkRung()
+        Log.w(tag, "PTT recognizer rung failed captureId=$captureId rung=$candidate: ${err.message}")
+      }
+    }
+    throw lastFailure ?: IllegalStateException("Speech recognizer unavailable")
+  }
+
+  private fun pushToTalkRecognizerIntent(rung: PushToTalkRecognitionRung): Intent =
+    Intent(RecognizerIntent.ACTION_RECOGNIZE_SPEECH).apply {
+      putExtra(RecognizerIntent.EXTRA_LANGUAGE_MODEL, RecognizerIntent.LANGUAGE_MODEL_FREE_FORM)
+      putExtra(RecognizerIntent.EXTRA_LANGUAGE, resolvedSpeechLocaleTag())
+      putExtra(RecognizerIntent.EXTRA_PARTIAL_RESULTS, true)
+      putExtra(RecognizerIntent.EXTRA_MAX_RESULTS, 3)
+      putExtra(RecognizerIntent.EXTRA_CALLING_PACKAGE, context.packageName)
+      when (rung) {
+        is PushToTalkRecognitionRung.RawAudioSegmented ->
+          if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.TIRAMISU) {
+            applyRawAudioSegmentedExtras(this, rung.source)
+          }
+        PushToTalkRecognitionRung.SilenceSegmented -> {
+          putExtra(RecognizerIntent.EXTRA_SPEECH_INPUT_COMPLETE_SILENCE_LENGTH_MILLIS, 2500)
+          putExtra(RecognizerIntent.EXTRA_SPEECH_INPUT_POSSIBLY_COMPLETE_SILENCE_LENGTH_MILLIS, 1800)
+          if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.TIRAMISU) {
+            applySilenceSegmentedExtras(this)
+          }
+        }
+        PushToTalkRecognitionRung.RestartingSingleSession -> {
+          putExtra(RecognizerIntent.EXTRA_SPEECH_INPUT_COMPLETE_SILENCE_LENGTH_MILLIS, 2500)
+          putExtra(RecognizerIntent.EXTRA_SPEECH_INPUT_POSSIBLY_COMPLETE_SILENCE_LENGTH_MILLIS, 1800)
+        }
+      }
+    }
+
+  // API 33 RecognizerIntent extras live behind @RequiresApi so min-SDK lint stays meaningful;
+  // segmented rungs are only ever constructed on TIRAMISU+ (see pushToTalkRecognitionCandidates).
+  @RequiresApi(Build.VERSION_CODES.TIRAMISU)
+  private fun applyRawAudioSegmentedExtras(
+    intent: Intent,
+    source: PushToTalkAudioSource,
+  ) {
+    intent.putExtra(RecognizerIntent.EXTRA_AUDIO_SOURCE, source.readDescriptor)
+    intent.putExtra(RecognizerIntent.EXTRA_AUDIO_SOURCE_CHANNEL_COUNT, 1)
+    intent.putExtra(RecognizerIntent.EXTRA_AUDIO_SOURCE_ENCODING, AudioFormat.ENCODING_PCM_16BIT)
+    intent.putExtra(RecognizerIntent.EXTRA_AUDIO_SOURCE_SAMPLING_RATE, pushToTalkSampleRateHz)
+    intent.putExtra(RecognizerIntent.EXTRA_SEGMENTED_SESSION, RecognizerIntent.EXTRA_AUDIO_SOURCE)
+  }
+
+  @RequiresApi(Build.VERSION_CODES.TIRAMISU)
+  private fun applySilenceSegmentedExtras(intent: Intent) {
+    intent.putExtra(
+      RecognizerIntent.EXTRA_SEGMENTED_SESSION,
+      RecognizerIntent.EXTRA_SPEECH_INPUT_COMPLETE_SILENCE_LENGTH_MILLIS,
+    )
+  }
+
+  @SuppressLint("MissingPermission")
+  private fun openPushToTalkAudioSource(): PushToTalkAudioSource {
+    check(Build.VERSION.SDK_INT >= Build.VERSION_CODES.TIRAMISU)
+    val minBufferSize =
+      AudioRecord.getMinBufferSize(
+        pushToTalkSampleRateHz,
+        AudioFormat.CHANNEL_IN_MONO,
+        AudioFormat.ENCODING_PCM_16BIT,
+      )
+    check(minBufferSize > 0) { "AudioRecord buffer unavailable" }
+
+    val pipe = ParcelFileDescriptor.createPipe()
+    var recorder: AudioRecord? = null
+    var writeStream: ParcelFileDescriptor.AutoCloseOutputStream? = null
+    try {
+      recorder =
+        AudioRecord(
+          MediaRecorder.AudioSource.VOICE_RECOGNITION,
+          pushToTalkSampleRateHz,
+          AudioFormat.CHANNEL_IN_MONO,
+          AudioFormat.ENCODING_PCM_16BIT,
+          minBufferSize * 2,
+        )
+      check(recorder.state == AudioRecord.STATE_INITIALIZED) { "AudioRecord initialization failed" }
+      recorder.startRecording()
+      check(recorder.recordingState == AudioRecord.RECORDSTATE_RECORDING) { "AudioRecord did not start" }
+      val activeRecorder = checkNotNull(recorder)
+      val activeWriteStream = ParcelFileDescriptor.AutoCloseOutputStream(pipe[1])
+      writeStream = activeWriteStream
+      val source = PushToTalkAudioSource(pipe[0], activeWriteStream, activeRecorder)
+      source.pumpJob =
+        gatewayWorkScope.launch(Dispatchers.IO) {
+          val buffer = ByteArray(minBufferSize.coerceAtLeast(4_096))
+          try {
+            while (currentCoroutineContext().isActive) {
+              val bytesRead = activeRecorder.read(buffer, 0, buffer.size)
+              if (bytesRead <= 0) break
+              _inputLevel.value =
+                TalkAudioLevel.smoothed(_inputLevel.value, TalkAudioLevel.pcm16Level(buffer, bytesRead))
+              activeWriteStream.write(buffer, 0, bytesRead)
+            }
+          } catch (err: IOException) {
+            Log.d(tag, "PTT audio pipe closed: ${err.message}")
+          } finally {
+            source.finishFromPump()
+          }
+        }
+      return source
+    } catch (err: Throwable) {
+      runCatching { recorder?.stop() }
+      runCatching { recorder?.release() }
+      runCatching { writeStream?.close() }
+      if (writeStream == null) runCatching { pipe[1].close() }
+      runCatching { pipe[0].close() }
+      throw err
+    }
+  }
+
+  private fun schedulePushToTalkRestart(
+    delayMs: Long,
+    advanceRung: Boolean,
+  ) {
+    val captureId = activePttCaptureId ?: return
+    if (pttReleaseCompletion != null) return
+    val rung = pttRecognitionRung ?: return
+    val firstCandidate =
+      when (rung) {
+        is PushToTalkRecognitionRung.RawAudioSegmented ->
+          if (advanceRung) {
+            PushToTalkRecognitionCandidate.SilenceSegmented
+          } else {
+            PushToTalkRecognitionCandidate.RawAudioSegmented
+          }
+        PushToTalkRecognitionRung.SilenceSegmented ->
+          if (advanceRung) {
+            PushToTalkRecognitionCandidate.RestartingSingleSession
+          } else {
+            PushToTalkRecognitionCandidate.SilenceSegmented
+          }
+        PushToTalkRecognitionRung.RestartingSingleSession ->
+          PushToTalkRecognitionCandidate.RestartingSingleSession
+      }
+    commitPushToTalkLivePartial()
+    closePushToTalkRung()
+    restartJob?.cancel()
+    restartJob =
+      gatewayWorkScope.launch {
+        delay(delayMs)
+        mainHandler.post {
+          if (activePttCaptureId != captureId || pttReleaseCompletion != null || stopRequested) return@post
+          try {
+            startPushToTalkRecognition(captureId, firstCandidate)
+          } catch (err: Throwable) {
+            _isListening.value = false
+            setTalkFailure(nativeText("Talk failed: \$message", err.message ?: err::class.simpleName.orEmpty()))
+          }
+        }
+      }
+  }
+
+  private fun closePushToTalkRung() {
+    (pttRecognitionRung as? PushToTalkRecognitionRung.RawAudioSegmented)?.source?.close()
+    pttRecognitionRung = null
+  }
+
+  private fun commitPushToTalkLivePartial() {
+    val partial = pttLivePartial.trim()
+    if (partial.isNotEmpty()) {
+      pttFinalSegments += partial
+    }
+    pttLivePartial = ""
+  }
+
  private fun startListeningInternal(markListening: Boolean) {
    val r = recognizer ?: return
    val intent =
@@ -2153,6 +2465,18 @@ class TalkModeManager internal constructor(
    isFinal: Boolean,
  ) {
    val trimmed = text.trim()
+    if (activePttCaptureId != null) {
+      if (trimmed.isNotEmpty()) {
+        if (isFinal) {
+          pttFinalSegments += trimmed
+          pttLivePartial = ""
+        } else {
+          pttLivePartial = trimmed
+        }
+        lastHeardAtMs = SystemClock.elapsedRealtime()
+      }
+      return
+    }
    if (_isSpeaking.value && interruptOnSpeech) {
      if (shouldInterrupt(trimmed)) {
        stopSpeaking()
@@ -2188,13 +2512,19 @@ class TalkModeManager internal constructor(

  private fun checkSilence(captureId: String) {
    if (!_isListening.value) return
-    val transcript = lastTranscript.trim()
+    val transcript =
+      if (activePttCaptureId != null) {
+        PushToTalkTranscriptMerger.merge(pttFinalSegments, pttLivePartial)
+      } else {
+        lastTranscript.trim()
+      }
    if (transcript.isEmpty()) return
    val lastHeard = lastHeardAtMs ?: return
    val elapsed = SystemClock.elapsedRealtime() - lastHeard
    if (elapsed < silenceWindowMs) return
    if (activePttCaptureId != null) {
      if (pttAutoStopEnabled) {
+        if (pttReleaseCompletion != null) return
        gatewayWorkScope.launch { endPushToTalk(captureId) }
      }
      return
@@ -2285,14 +2615,78 @@ class TalkModeManager internal constructor(
    }
  }

+  private suspend fun awaitPushToTalkRelease(captureId: String) {
+    if (activePttCaptureId != captureId) return
+    restartJob?.cancel()
+    restartJob = null
+    val rung = pttRecognitionRung ?: return
+    // onResults, onError, and onEndOfSegmentedSession normally arrive well under a second,
+    // so typical release latency is unchanged. The five-second bound only caps pathological recognizers;
+    // leaving early truncates final words, which is worse than waiting.
+    pttReleaseCompletion?.let { existing ->
+      awaitPushToTalkReleaseCompletion(existing, pushToTalkReleaseGraceMs)
+      return
+    }
+    if (!_isListening.value || recognizer == null) return
+
+    val completion = CompletableDeferred<Unit>()
+    pttReleaseCompletion = completion
+    _isListening.value = false
+    _inputLevel.value = 0f
+    when (rung) {
+      is PushToTalkRecognitionRung.RawAudioSegmented -> {
+        rung.source.requestFinish()
+        // EXTRA_AUDIO_SOURCE is optional: a service may ignore the pipe and run its own mic,
+        // so closing our AudioRecord alone would leave it listening past release. stopListening
+        // forces its endpointer; for pipe-consuming services it is redundant after EOF.
+        runCatching { recognizer?.stopListening() }.onFailure { completion.complete(Unit) }
+      }
+      PushToTalkRecognitionRung.SilenceSegmented,
+      PushToTalkRecognitionRung.RestartingSingleSession,
+      -> runCatching { recognizer?.stopListening() }.onFailure { completion.complete(Unit) }
+    }
+    awaitPushToTalkReleaseCompletion(completion, pushToTalkReleaseGraceMs)
+    if (pttReleaseCompletion === completion) {
+      pttReleaseCompletion = null
+    }
+  }
+
+  private suspend fun drainPushToTalkReleaseBeforeBegin() {
+    val deadline = SystemClock.elapsedRealtime() + pushToTalkReleaseDrainTimeoutMs
+    while (true) {
+      val remainingMs = deadline - SystemClock.elapsedRealtime()
+      val release = pttReleaseCompletion
+      if (release != null && remainingMs > 0) {
+        awaitPushToTalkReleaseCompletion(release, remainingMs)
+      }
+      if (activePttCaptureId == null && pttReleaseCompletion == null) return
+      if (SystemClock.elapsedRealtime() >= deadline) return
+      yield()
+    }
+  }
+
+  private suspend fun awaitPushToTalkReleaseCompletion(
+    completion: CompletableDeferred<Unit>,
+    timeoutMs: Long,
+  ) {
+    try {
+      withTimeoutOrNull(timeoutMs) { completion.await() }
+    } catch (err: CancellationException) {
+      if (completion.isCancelled && currentCoroutineContext().isActive) return
+      throw err
+    }
+  }
+
  private fun clearPushToTalkRecognition(captureId: String): ClearedPushToTalkCapture? {
    if (activePttCaptureId != captureId) return null
-    val transcript = lastTranscript.trim()
+    val transcript = PushToTalkTranscriptMerger.merge(pttFinalSegments, pttLivePartial)
    val completion = pttCompletion
    pttTimeoutJob?.cancel()
    pttTimeoutJob = null
    pttAutoStopEnabled = false
    pttCompletion = null
+    pttReleaseCompletion?.cancel()
+    pttReleaseCompletion = null
    activePttCaptureId = null
    _isListening.value = false
    listeningMode = false
@@ -2300,8 +2694,12 @@ class TalkModeManager internal constructor(
    recognizer?.cancel()
    recognizer?.destroy()
    recognizer = null
+    closePushToTalkRung()
+    pttFinalSegments.clear()
+    pttLivePartial = ""
    lastTranscript = ""
    lastHeardAtMs = null
+    _inputLevel.value = 0f
    return ClearedPushToTalkCapture(transcript = transcript, completion = completion)
  }

@@ -2966,24 +3364,39 @@ class TalkModeManager internal constructor(
    }
  }

-  private val listener =
+  private val listener = recognitionListener(captureId = null)
+
+  private fun recognitionListener(captureId: String?): RecognitionListener =
    object : RecognitionListener {
      override fun onReadyForSpeech(params: Bundle?) {
+        if (!acceptRecognitionCallback(captureId)) return
        // Only a live listening session may claim the status; a speech-interrupt
        // recognizer readying during playback must not touch Thinking state.
-        if (_isEnabled.value && _isListening.value) {
+        if (activePttCaptureId != null && _isListening.value) {
+          setStatus(nativeText("Listening (PTT)"))
+        } else if (_isEnabled.value && _isListening.value) {
          setStatus(nativeText("Listening"))
        }
      }

-      override fun onBeginningOfSpeech() {}
+      override fun onBeginningOfSpeech() {
+        if (!acceptRecognitionCallback(captureId)) return
+      }

-      override fun onRmsChanged(rmsdB: Float) {}
+      override fun onRmsChanged(rmsdB: Float) {
+        if (!acceptRecognitionCallback(captureId)) return
+        if (activePttCaptureId != null && pttRecognitionRung !is PushToTalkRecognitionRung.RawAudioSegmented) {
+          _inputLevel.value = ((rmsdB + 2f) / 12f).coerceIn(0f, 1f)
+        }
+      }

      override fun onBufferReceived(buffer: ByteArray?) {}

      override fun onEndOfSpeech() {
+        if (!acceptRecognitionCallback(captureId)) return
        clearListenWatchdog()
+        _inputLevel.value = 0f
+        if (activePttCaptureId != null) return
        // Don't restart while a transcript is being processed — the recognizer
        // competing for audio resources kills AudioTrack PCM playback.
        if (!finalizeInFlight) {
@@ -2992,8 +3405,38 @@ class TalkModeManager internal constructor(
      }

      override fun onError(error: Int) {
+        if (!acceptRecognitionCallback(captureId)) return
        if (stopRequested) return
        _isListening.value = false
+        _inputLevel.value = 0f
+        if (activePttCaptureId != null) {
+          pttReleaseCompletion?.let {
+            it.complete(Unit)
+            return
+          }
+          if (error == SpeechRecognizer.ERROR_INSUFFICIENT_PERMISSIONS) {
+            setStatus(nativeText("Microphone permission required"))
+            return
+          }
+          setStatus(
+            when (error) {
+              SpeechRecognizer.ERROR_AUDIO -> nativeText("Audio error")
+              SpeechRecognizer.ERROR_CLIENT -> nativeText("Client error")
+              SpeechRecognizer.ERROR_NETWORK -> nativeText("Network error")
+              SpeechRecognizer.ERROR_NETWORK_TIMEOUT -> nativeText("Network timeout")
+              SpeechRecognizer.ERROR_NO_MATCH -> nativeText("Listening (PTT)")
+              SpeechRecognizer.ERROR_RECOGNIZER_BUSY -> nativeText("Recognizer busy")
+              SpeechRecognizer.ERROR_SERVER -> nativeText("Server error")
+              SpeechRecognizer.ERROR_SPEECH_TIMEOUT -> nativeText("Listening (PTT)")
+              else -> nativeText("Speech error (\$error)", error)
+            },
+          )
+          schedulePushToTalkRestart(
+            delayMs = 600L,
+            advanceRung = pttRecognitionRung !is PushToTalkRecognitionRung.RestartingSingleSession,
+          )
+          return
+        }
        if (error == SpeechRecognizer.ERROR_INSUFFICIENT_PERMISSIONS) {
          setStatus(nativeText("Microphone permission required"))
          return
@@ -3016,21 +3459,60 @@ class TalkModeManager internal constructor(
      }

      override fun onResults(results: Bundle?) {
+        if (!acceptRecognitionCallback(captureId)) return
        val list = results?.getStringArrayList(SpeechRecognizer.RESULTS_RECOGNITION).orEmpty()
        list.firstOrNull()?.let { handleTranscript(it, isFinal = true) }
+        if (activePttCaptureId != null) {
+          _isListening.value = false
+          _inputLevel.value = 0f
+          pttReleaseCompletion?.let {
+            it.complete(Unit)
+            return
+          }
+          schedulePushToTalkRestart(
+            delayMs = pushToTalkRestartDelayMs,
+            advanceRung = pttRecognitionRung !is PushToTalkRecognitionRung.RestartingSingleSession,
+          )
+          return
+        }
        scheduleRestart()
      }

      override fun onPartialResults(partialResults: Bundle?) {
+        if (!acceptRecognitionCallback(captureId)) return
        val list = partialResults?.getStringArrayList(SpeechRecognizer.RESULTS_RECOGNITION).orEmpty()
        list.firstOrNull()?.let { handleTranscript(it, isFinal = false) }
      }

+      override fun onSegmentResults(segmentResults: Bundle) {
+        if (!acceptRecognitionCallback(captureId) || activePttCaptureId == null) return
+        val list = segmentResults.getStringArrayList(SpeechRecognizer.RESULTS_RECOGNITION).orEmpty()
+        list.firstOrNull()?.let { handleTranscript(it, isFinal = true) }
+      }
+
+      override fun onEndOfSegmentedSession() {
+        if (!acceptRecognitionCallback(captureId) || activePttCaptureId == null) return
+        _isListening.value = false
+        _inputLevel.value = 0f
+        pttReleaseCompletion?.let {
+          it.complete(Unit)
+          return
+        }
+        schedulePushToTalkRestart(
+          delayMs = 180L,
+          advanceRung = shouldAdvancePushToTalkRungAfterSegmentedSession(pttRecognitionRung?.candidate ?: return),
+        )
+      }
+
      override fun onEvent(
        eventType: Int,
        params: Bundle?,
      ) {}
    }
+
+  // SpeechRecognizer can post callbacks after destroy. Binding each listener to its
+  // capture prevents a retired session from mutating normal Talk or the next PTT hold.
+  private fun acceptRecognitionCallback(captureId: String?): Boolean = captureId == activePttCaptureId
 }

 private fun JsonElement?.asObjectOrNull(): JsonObject? = this as? JsonObject
@@ -0,0 +1,56 @@
+package ai.openclaw.app.voice
+
+import org.junit.Assert.assertEquals
+import org.junit.Test
+
+class PushToTalkRecognitionLadderTest {
+  @Test
+  fun api33StartsWithRawAudioThenFallsThroughInOrder() {
+    assertEquals(
+      listOf(
+        PushToTalkRecognitionCandidate.RawAudioSegmented,
+        PushToTalkRecognitionCandidate.SilenceSegmented,
+        PushToTalkRecognitionCandidate.RestartingSingleSession,
+      ),
+      pushToTalkRecognitionCandidates(supportsSegmentedRecognition = true, first = null),
+    )
+  }
+
+  @Test
+  fun olderApisUseRestartingSingleSessionOnly() {
+    assertEquals(
+      listOf(PushToTalkRecognitionCandidate.RestartingSingleSession),
+      pushToTalkRecognitionCandidates(supportsSegmentedRecognition = false, first = null),
+    )
+  }
+
+  @Test
+  fun degradedHoldNeverClimbsBackUpTheLadder() {
+    assertEquals(
+      listOf(
+        PushToTalkRecognitionCandidate.SilenceSegmented,
+        PushToTalkRecognitionCandidate.RestartingSingleSession,
+      ),
+      pushToTalkRecognitionCandidates(
+        supportsSegmentedRecognition = true,
+        first = PushToTalkRecognitionCandidate.SilenceSegmented,
+      ),
+    )
+  }
+
+  @Test
+  fun rawSegmentedSessionEndAdvancesToSilenceButSilenceSessionEndRearms() {
+    assertEquals(
+      true,
+      shouldAdvancePushToTalkRungAfterSegmentedSession(
+        PushToTalkRecognitionCandidate.RawAudioSegmented,
+      ),
+    )
+    assertEquals(
+      false,
+      shouldAdvancePushToTalkRungAfterSegmentedSession(
+        PushToTalkRecognitionCandidate.SilenceSegmented,
+      ),
+    )
+  }
+}
@@ -0,0 +1,90 @@
+package ai.openclaw.app.voice
+
+import org.junit.Assert.assertEquals
+import org.junit.Test
+
+class PushToTalkTranscriptMergerTest {
+  @Test
+  fun joinsMultipleFinalSegmentsInOrder() {
+    assertEquals(
+      "first thought. second thought. third thought",
+      PushToTalkTranscriptMerger.merge(
+        finalSegments = listOf("first thought", "second thought", "third thought"),
+        livePartial = null,
+      ),
+    )
+  }
+
+  @Test
+  fun preservesTerminalPunctuationWithoutDoublingIt() {
+    assertEquals(
+      "ready? yes! \"done.\" next",
+      PushToTalkTranscriptMerger.merge(
+        finalSegments = listOf("ready?", "yes!", "\"done.\"", "next"),
+        livePartial = null,
+      ),
+    )
+  }
+
+  @Test
+  fun preservesLocalePunctuationWithoutInjectingAsciiSeparators() {
+    assertEquals(
+      "你好。 世界",
+      PushToTalkTranscriptMerger.merge(
+        finalSegments = listOf("你好。", "世界"),
+        livePartial = null,
+      ),
+    )
+    assertEquals(
+      "هل أنت جاهز؟ نعم",
+      PushToTalkTranscriptMerger.merge(
+        finalSegments = listOf("هل أنت جاهز؟", "نعم"),
+        livePartial = null,
+      ),
+    )
+  }
+
+  @Test
+  fun omitsTrailingPartialThatDuplicatesLastFinal() {
+    assertEquals(
+      "first. SAME words",
+      PushToTalkTranscriptMerger.merge(
+        finalSegments = listOf("first", "SAME words"),
+        livePartial = "  same   WORDS  ",
+      ),
+    )
+  }
+
+  @Test
+  fun appendsDistinctTrailingPartial() {
+    assertEquals(
+      "first final. trailing words",
+      PushToTalkTranscriptMerger.merge(
+        finalSegments = listOf("first final"),
+        livePartial = "trailing words",
+      ),
+    )
+  }
+
+  @Test
+  fun returnsPartialForPartialOnlyHold() {
+    assertEquals(
+      "unfinished thought",
+      PushToTalkTranscriptMerger.merge(
+        finalSegments = emptyList(),
+        livePartial = " unfinished thought ",
+      ),
+    )
+  }
+
+  @Test
+  fun returnsEmptyForEmptyResults() {
+    assertEquals(
+      "",
+      PushToTalkTranscriptMerger.merge(
+        finalSegments = listOf("", "   "),
+        livePartial = "  ",
+      ),
+    )
+  }
+}
@@ -11,8 +11,11 @@ import ai.openclaw.app.i18n.verbatimText
 import android.Manifest
 import android.content.ComponentName
 import android.content.IntentFilter
+import android.os.Bundle
 import android.os.SystemClock
+import android.speech.RecognitionListener
 import android.speech.RecognitionService
+import android.speech.SpeechRecognizer
 import kotlinx.coroutines.CompletableDeferred
 import kotlinx.coroutines.CoroutineDispatcher
 import kotlinx.coroutines.CoroutineScope
@@ -26,6 +29,7 @@ import kotlinx.coroutines.async
 import kotlinx.coroutines.flow.MutableStateFlow
 import kotlinx.coroutines.launch
 import kotlinx.coroutines.test.StandardTestDispatcher
+import kotlinx.coroutines.test.advanceTimeBy
 import kotlinx.coroutines.test.advanceUntilIdle
 import kotlinx.coroutines.test.currentTime
 import kotlinx.coroutines.test.resetMain
@@ -279,6 +283,147 @@ class TalkModeManagerTest {
      }
    }

+  @Test
+  fun segmentDuringPushToTalkReleaseWaitsForEndOfSegmentedSession() {
+    val manager = createManager()
+    val releaseCompletion = CompletableDeferred<Unit>()
+    setPrivateField(manager, "activePttCaptureId", "capture-1")
+    setPrivateField(manager, "pttReleaseCompletion", releaseCompletion)
+    val listener = recognitionListener(manager, "capture-1")
+    val segment =
+      Bundle().apply {
+        putStringArrayList(SpeechRecognizer.RESULTS_RECOGNITION, arrayListOf("first segment"))
+      }
+
+    listener.onSegmentResults(segment)
+
+    assertFalse(releaseCompletion.isCompleted)
+    assertEquals(listOf("first segment"), readPrivateField(manager, "pttFinalSegments"))
+
+    listener.onEndOfSegmentedSession()
+
+    assertTrue(releaseCompletion.isCompleted)
+  }
+
+  @Test
+  @OptIn(ExperimentalCoroutinesApi::class)
+  fun releaseKeepsWaitingPastOldGraceForLateTerminalSegment() =
+    runTest {
+      val manager = createManager(isConnected = { false })
+      val releaseCompletion = CompletableDeferred<Unit>()
+      setPrivateField(manager, "activePttCaptureId", "capture-1")
+      setPrivateField(manager, "pttReleaseCompletion", releaseCompletion)
+      setPrivateField(manager, "pttRecognitionRung", silenceSegmentedRung())
+      @Suppress("UNCHECKED_CAST")
+      (readPrivateField(manager, "pttFinalSegments") as MutableList<String>) += "early segment"
+      val listener = recognitionListener(manager, "capture-1")
+      Dispatchers.setMain(StandardTestDispatcher(testScheduler))
+      try {
+        val ending = async { manager.endPushToTalk("capture-1") }
+        runCurrent()
+
+        advanceTimeBy(1_200)
+        listener.onSegmentResults(
+          Bundle().apply {
+            putStringArrayList(SpeechRecognizer.RESULTS_RECOGNITION, arrayListOf("late segment"))
+          },
+        )
+        assertFalse(ending.isCompleted)
+
+        listener.onEndOfSegmentedSession()
+        advanceUntilIdle()
+
+        assertEquals("early segment. late segment", ending.await().transcript)
+      } finally {
+        Dispatchers.resetMain()
+      }
+    }
+
+  @Test
+  @OptIn(ExperimentalCoroutinesApi::class)
+  fun cancelledEndPushToTalkClearsPendingReleaseBeforeNextBegin() =
+    runTest {
+      val app = RuntimeEnvironment.getApplication()
+      shadowOf(app).grantPermissions(Manifest.permission.RECORD_AUDIO)
+      val packageManager = shadowOf(app.packageManager)
+      val speechService = ComponentName(app, "TestSpeechRecognitionService")
+      packageManager.addServiceIfNotPresent(speechService)
+      packageManager.addIntentFilterForService(speechService, IntentFilter(RecognitionService.SERVICE_INTERFACE))
+      val manager = createManager()
+      setPrivateField(manager, "activePttCaptureId", "capture-a")
+      setPrivateField(manager, "pttReleaseCompletion", CompletableDeferred<Unit>())
+      setPrivateField(manager, "pttRecognitionRung", silenceSegmentedRung())
+      @Suppress("UNCHECKED_CAST")
+      (readPrivateField(manager, "pttFinalSegments") as MutableList<String>) += "capture a"
+      Dispatchers.setMain(StandardTestDispatcher(testScheduler))
+      try {
+        val ending = async { manager.endPushToTalk("capture-a") }
+        runCurrent()
+        ending.cancel()
+        runCurrent()
+        ending.join()
+
+        assertTrue(ending.isCancelled)
+        assertNull(readPrivateField(manager, "activePttCaptureId"))
+        assertNull(readPrivateField(manager, "pttReleaseCompletion"))
+        assertEquals(emptyList<String>(), readPrivateField(manager, "pttFinalSegments"))
+
+        val started = manager.beginPushToTalk(allowNewCapture = true)
+
+        assertEquals(started.captureId, readPrivateField(manager, "activePttCaptureId"))
+        assertEquals(emptyList<String>(), readPrivateField(manager, "pttFinalSegments"))
+      } finally {
+        manager.stopAllCapture()
+        Dispatchers.resetMain()
+      }
+    }
+
+  @Test
+  @OptIn(ExperimentalCoroutinesApi::class)
+  fun replacementBeginDrainsPendingReleaseBeforeStartingNewCapture() =
+    runTest {
+      val app = RuntimeEnvironment.getApplication()
+      shadowOf(app).grantPermissions(Manifest.permission.RECORD_AUDIO)
+      val packageManager = shadowOf(app.packageManager)
+      val speechService = ComponentName(app, "TestSpeechRecognitionService")
+      packageManager.addServiceIfNotPresent(speechService)
+      packageManager.addIntentFilterForService(speechService, IntentFilter(RecognitionService.SERVICE_INTERFACE))
+      var connectionChecks = 0
+      val manager =
+        createManager(
+          isConnected = {
+            connectionChecks += 1
+            connectionChecks != 2
+          },
+        )
+      val releaseCompletion = CompletableDeferred<Unit>()
+      setPrivateField(manager, "activePttCaptureId", "capture-a")
+      setPrivateField(manager, "pttReleaseCompletion", releaseCompletion)
+      setPrivateField(manager, "pttRecognitionRung", silenceSegmentedRung())
+      @Suppress("UNCHECKED_CAST")
+      (readPrivateField(manager, "pttFinalSegments") as MutableList<String>) += "first segment"
+      Dispatchers.setMain(StandardTestDispatcher(testScheduler))
+      try {
+        val ending = async { manager.endPushToTalk("capture-a") }
+        runCurrent()
+        val starting = async { manager.beginPushToTalk(allowNewCapture = true) }
+        runCurrent()
+
+        releaseCompletion.complete(Unit)
+        advanceUntilIdle()
+
+        val ended = ending.await()
+        val started = starting.await()
+        assertEquals("offline", ended.status)
+        assertEquals("first segment", ended.transcript)
+        assertEquals(started.captureId, readPrivateField(manager, "activePttCaptureId"))
+        assertEquals(emptyList<String>(), readPrivateField(manager, "pttFinalSegments"))
+      } finally {
+        manager.stopAllCapture()
+        Dispatchers.resetMain()
+      }
+    }
+
  @Test
  fun duplicateFinalForPendingTalkRunDoesNotStartAllResponseTts() {
    val manager = createManager()
@@ -1125,6 +1270,20 @@ class TalkModeManagerTest {
    return method.invoke(manager, length) as Boolean
  }

+  private fun recognitionListener(
+    manager: TalkModeManager,
+    captureId: String,
+  ): RecognitionListener {
+    val method = manager.javaClass.getDeclaredMethod("recognitionListener", String::class.java)
+    method.isAccessible = true
+    return method.invoke(manager, captureId) as RecognitionListener
+  }
+
+  private fun silenceSegmentedRung(): Any {
+    val clazz = Class.forName("ai.openclaw.app.voice.PushToTalkRecognitionRung\$SilenceSegmented")
+    return requireNotNull(clazz.getField("INSTANCE").get(null))
+  }
+
  private fun chatFinalPayload(
    runId: String,
    text: String,