talk: add configurable silence timeout

Author: danodoesdesign

CHANGELOG.md

@@ -8,6 +8,7 @@ Docs: https://docs.openclaw.ai
 
 - TUI: infer the active agent from the current workspace when launched inside a configured agent workspace, while preserving explicit `agent:` session targets. (#39591) thanks @arceus77-7.
 - Tools/Brave web search: add opt-in `tools.web.search.brave.mode: "llm-context"` so `web_search` can call Brave's LLM Context endpoint and return extracted grounding snippets with source metadata, plus config/docs/test coverage. (#33383) Thanks @thirumaleshp.
+- Talk mode: add top-level `talk.silenceTimeoutMs` config so Talk waits a configurable amount of silence before auto-sending the current transcript, while keeping each platform's existing default pause window when unset. (#39607) Thanks @danodoesdesign. Fixes #17147.
 
 ### Fixes
 

apps/android/app/src/main/java/ai/openclaw/app/voice/TalkModeManager.kt

@@ -59,8 +59,8 @@ class TalkModeManager(
     private const val tag = "TalkMode"
     private const val defaultModelIdFallback = "eleven_v3"
     private const val defaultOutputFormatFallback = "pcm_24000"
-private const val defaultTalkProvider = "elevenlabs"
-    private const val silenceWindowMs = 500L
+    private const val defaultTalkProvider = "elevenlabs"
+    private const val defaultSilenceTimeoutMs = 700L
     private const val listenWatchdogMs = 12_000L
     private const val chatFinalWaitWithSubscribeMs = 45_000L
     private const val chatFinalWaitWithoutSubscribeMs = 6_000L
@@ -105,6 +105,14 @@ private const val defaultTalkProvider = "elevenlabs"
         normalizedPayload = false,
       )
     }
+
+    internal fun resolvedSilenceTimeoutMs(talk: JsonObject?): Long {
+      val timeout = talk?.get("silenceTimeoutMs").asDoubleOrNull() ?: return defaultSilenceTimeoutMs
+      if (timeout <= 0 || timeout % 1.0 != 0.0 || timeout > Long.MAX_VALUE.toDouble()) {
+        return defaultSilenceTimeoutMs
+      }
+      return timeout.toLong()
+    }
   }
 
   private val mainHandler = Handler(Looper.getMainLooper())
@@ -134,7 +142,7 @@ private const val defaultTalkProvider = "elevenlabs"
   private var listeningMode = false
 
   private var silenceJob: Job? = null
-  private val silenceWindowMs = 700L
+  private var silenceWindowMs = defaultSilenceTimeoutMs
   private var lastTranscript: String = ""
   private var lastHeardAtMs: Long? = null
   private var lastSpokenText: String? = null
@@ -1411,6 +1419,7 @@ private const val defaultTalkProvider = "elevenlabs"
         activeConfig?.get("outputFormat")?.asStringOrNull()?.trim()?.takeIf { it.isNotEmpty() }
       val key = activeConfig?.get("apiKey")?.asStringOrNull()?.trim()?.takeIf { it.isNotEmpty() }
       val interrupt = talk?.get("interruptOnSpeech")?.asBooleanOrNull()
+      val silenceTimeoutMs = resolvedSilenceTimeoutMs(talk)
 
       if (!isCanonicalMainSessionKey(mainSessionKey)) {
         mainSessionKey = mainKey
@@ -1427,7 +1436,11 @@ private const val defaultTalkProvider = "elevenlabs"
       if (!modelOverrideActive) currentModelId = defaultModelId
       defaultOutputFormat = outputFormat ?: defaultOutputFormatFallback
       apiKey = key ?: envKey?.takeIf { it.isNotEmpty() }
-      Log.d(tag, "reloadConfig apiKey=${if (apiKey != null) "set" else "null"} voiceId=$defaultVoiceId")
+      silenceWindowMs = silenceTimeoutMs
+      Log.d(
+        tag,
+        "reloadConfig apiKey=${if (apiKey != null) "set" else "null"} voiceId=$defaultVoiceId silenceTimeoutMs=$silenceTimeoutMs",
+      )
       if (interrupt != null) interruptOnSpeech = interrupt
       activeProviderIsElevenLabs = activeProvider == defaultTalkProvider
       if (!activeProviderIsElevenLabs) {
@@ -1441,6 +1454,7 @@ private const val defaultTalkProvider = "elevenlabs"
       }
       configLoaded = true
     } catch (_: Throwable) {
+      silenceWindowMs = defaultSilenceTimeoutMs
       defaultVoiceId = envVoice?.takeIf { it.isNotEmpty() } ?: sagVoice?.takeIf { it.isNotEmpty() }
       defaultModelId = defaultModelIdFallback
       if (!modelOverrideActive) currentModelId = defaultModelId

apps/android/app/src/test/java/ai/openclaw/app/voice/TalkModeConfigParsingTest.kt

@@ -54,4 +54,23 @@ class TalkModeConfigParsingTest {
     assertEquals("voice-legacy", selection?.config?.get("voiceId")?.jsonPrimitive?.content)
     assertEquals("legacy-key", selection?.config?.get("apiKey")?.jsonPrimitive?.content)
   }
+
+  @Test
+  fun readsConfiguredSilenceTimeoutMs() {
+    val talk = buildJsonObject { put("silenceTimeoutMs", 1500) }
+
+    assertEquals(1500L, TalkModeManager.resolvedSilenceTimeoutMs(talk))
+  }
+
+  @Test
+  fun defaultsSilenceTimeoutMsWhenMissing() {
+    assertEquals(700L, TalkModeManager.resolvedSilenceTimeoutMs(null))
+  }
+
+  @Test
+  fun defaultsSilenceTimeoutMsWhenInvalid() {
+    val talk = buildJsonObject { put("silenceTimeoutMs", 0) }
+
+    assertEquals(700L, TalkModeManager.resolvedSilenceTimeoutMs(talk))
+  }
 }

apps/ios/Sources/Voice/TalkModeManager.swift

@@ -34,6 +34,7 @@ final class TalkModeManager: NSObject {
     private typealias SpeechRequest = SFSpeechAudioBufferRecognitionRequest
     private static let defaultModelIdFallback = "eleven_v3"
     private static let defaultTalkProvider = "elevenlabs"
+    private static let defaultSilenceTimeoutMs = 900
     private static let redactedConfigSentinel = "__OPENCLAW_REDACTED__"
     var isEnabled: Bool = false
     var isListening: Bool = false
@@ -97,7 +98,7 @@ final class TalkModeManager: NSObject {
 
     private var gateway: GatewayNodeSession?
     private var gatewayConnected = false
-    private let silenceWindow: TimeInterval = 0.9
+    private var silenceWindow: TimeInterval = TimeInterval(Self.defaultSilenceTimeoutMs) / 1000
     private var lastAudioActivity: Date?
     private var noiseFloorSamples: [Double] = []
     private var noiseFloor: Double?
@@ -2001,6 +2002,24 @@ extension TalkModeManager {
             config: normalizedProviders[providerID] ?? [:])
     }
 
+    static func resolvedSilenceTimeoutMs(_ talk: [String: Any]?) -> Int {
+        switch talk?["silenceTimeoutMs"] {
+        case let timeout as Int where timeout > 0:
+            return timeout
+        case let timeout as Double
+            where timeout > 0 && timeout.rounded(.towardZero) == timeout && timeout <= Double(Int.max):
+            return Int(timeout)
+        case let timeout as NSNumber:
+            let value = timeout.doubleValue
+            if value > 0 && value.rounded(.towardZero) == value && value <= Double(Int.max) {
+                return Int(value)
+            }
+            return Self.defaultSilenceTimeoutMs
+        default:
+            return Self.defaultSilenceTimeoutMs
+        }
+    }
+
     func reloadConfig() async {
         guard let gateway else { return }
         self.pcmFormatUnavailable = false
@@ -2020,6 +2039,7 @@ extension TalkModeManager {
             }
             let activeProvider = selection?.provider ?? Self.defaultTalkProvider
             let activeConfig = selection?.config
+            let silenceTimeoutMs = Self.resolvedSilenceTimeoutMs(talk)
             self.defaultVoiceId = (activeConfig?["voiceId"] as? String)?
                 .trimmingCharacters(in: .whitespacesAndNewlines)
             if let aliases = activeConfig?["voiceAliases"] as? [String: Any] {
@@ -2067,8 +2087,9 @@ extension TalkModeManager {
             if let interrupt = talk?["interruptOnSpeech"] as? Bool {
                 self.interruptOnSpeech = interrupt
             }
+            self.silenceWindow = TimeInterval(silenceTimeoutMs) / 1000
             if selection != nil {
-                GatewayDiagnostics.log("talk config provider=\(activeProvider)")
+                GatewayDiagnostics.log("talk config provider=\(activeProvider) silenceTimeoutMs=\(silenceTimeoutMs)")
             }
         } catch {
             self.defaultModelId = Self.defaultModelIdFallback
@@ -2079,6 +2100,7 @@ extension TalkModeManager {
             self.gatewayTalkDefaultModelId = nil
             self.gatewayTalkApiKeyConfigured = false
             self.gatewayTalkConfigLoaded = false
+            self.silenceWindow = TimeInterval(Self.defaultSilenceTimeoutMs) / 1000
         }
     }
 

apps/ios/Tests/TalkModeConfigParsingTests.swift

@@ -47,4 +47,24 @@ import Testing
             userInfo: [NSLocalizedDescriptionKey: "queue enqueue failed"])
         #expect(TalkModeManager._test_isPCMFormatRejectedByAPI(error) == false)
     }
+
+    @Test func readsConfiguredSilenceTimeoutMs() {
+        let talk: [String: Any] = [
+            "silenceTimeoutMs": 1500,
+        ]
+
+        #expect(TalkModeManager.resolvedSilenceTimeoutMs(talk) == 1500)
+    }
+
+    @Test func defaultsSilenceTimeoutMsWhenMissing() {
+        #expect(TalkModeManager.resolvedSilenceTimeoutMs(nil) == 900)
+    }
+
+    @Test func defaultsSilenceTimeoutMsWhenInvalid() {
+        let talk: [String: Any] = [
+            "silenceTimeoutMs": 0,
+        ]
+
+        #expect(TalkModeManager.resolvedSilenceTimeoutMs(talk) == 900)
+    }
 }

apps/macos/Sources/OpenClaw/TalkModeRuntime.swift

@@ -12,6 +12,7 @@ actor TalkModeRuntime {
     private let ttsLogger = Logger(subsystem: "ai.openclaw", category: "talk.tts")
     private static let defaultModelIdFallback = "eleven_v3"
     private static let defaultTalkProvider = "elevenlabs"
+    private static let defaultSilenceTimeoutMs = 700
 
     private final class RMSMeter: @unchecked Sendable {
         private let lock = NSLock()
@@ -66,7 +67,7 @@ actor TalkModeRuntime {
     private var fallbackVoiceId: String?
     private var lastPlaybackWasPCM: Bool = false
 
-    private let silenceWindow: TimeInterval = 0.7
+    private var silenceWindow: TimeInterval = TimeInterval(TalkModeRuntime.defaultSilenceTimeoutMs) / 1000
     private let minSpeechRMS: Double = 1e-3
     private let speechBoostFactor: Double = 6.0
 
@@ -783,6 +784,7 @@ extension TalkModeRuntime {
         }
         self.defaultOutputFormat = cfg.outputFormat
         self.interruptOnSpeech = cfg.interruptOnSpeech
+        self.silenceWindow = TimeInterval(cfg.silenceTimeoutMs) / 1000
         self.apiKey = cfg.apiKey
         let hasApiKey = (cfg.apiKey?.isEmpty == false)
         let voiceLabel = (cfg.voiceId?.isEmpty == false) ? cfg.voiceId! : "none"
@@ -792,7 +794,8 @@ extension TalkModeRuntime {
                 "talk config voiceId=\(voiceLabel, privacy: .public) " +
                     "modelId=\(modelLabel, privacy: .public) " +
                     "apiKey=\(hasApiKey, privacy: .public) " +
-                    "interrupt=\(cfg.interruptOnSpeech, privacy: .public)")
+                    "interrupt=\(cfg.interruptOnSpeech, privacy: .public) " +
+                    "silenceTimeoutMs=\(cfg.silenceTimeoutMs, privacy: .public)")
     }
 
     private struct TalkRuntimeConfig {
@@ -801,6 +804,7 @@ extension TalkModeRuntime {
         let modelId: String?
         let outputFormat: String?
         let interruptOnSpeech: Bool
+        let silenceTimeoutMs: Int
         let apiKey: String?
     }
 
@@ -880,6 +884,21 @@ extension TalkModeRuntime {
             normalizedPayload: false)
     }
 
+    static func resolvedSilenceTimeoutMs(_ talk: [String: AnyCodable]?) -> Int {
+        if let timeout = talk?["silenceTimeoutMs"]?.intValue, timeout > 0 {
+            return timeout
+        }
+        if
+            let timeout = talk?["silenceTimeoutMs"]?.doubleValue,
+            timeout > 0,
+            timeout.rounded(.towardZero) == timeout,
+            timeout <= Double(Int.max)
+        {
+            return Int(timeout)
+        }
+        return Self.defaultSilenceTimeoutMs
+    }
+
     private func fetchTalkConfig() async -> TalkRuntimeConfig {
         let env = ProcessInfo.processInfo.environment
         let envVoice = env["ELEVENLABS_VOICE_ID"]?.trimmingCharacters(in: .whitespacesAndNewlines)
@@ -895,6 +914,7 @@ extension TalkModeRuntime {
             let selection = Self.selectTalkProviderConfig(talk)
             let activeProvider = selection?.provider ?? Self.defaultTalkProvider
             let activeConfig = selection?.config
+            let silenceTimeoutMs = Self.resolvedSilenceTimeoutMs(talk)
             let ui = snap.config?["ui"]?.dictionaryValue
             let rawSeam = ui?["seamColor"]?.stringValue?.trimmingCharacters(in: .whitespacesAndNewlines) ?? ""
             await MainActor.run {
@@ -939,6 +959,7 @@ extension TalkModeRuntime {
                 modelId: resolvedModel,
                 outputFormat: outputFormat,
                 interruptOnSpeech: interrupt ?? true,
+                silenceTimeoutMs: silenceTimeoutMs,
                 apiKey: resolvedApiKey)
         } catch {
             let resolvedVoice =
@@ -951,6 +972,7 @@ extension TalkModeRuntime {
                 modelId: Self.defaultModelIdFallback,
                 outputFormat: nil,
                 interruptOnSpeech: true,
+                silenceTimeoutMs: Self.defaultSilenceTimeoutMs,
                 apiKey: resolvedApiKey)
         }
     }

apps/macos/Tests/OpenClawIPCTests/TalkModeConfigParsingTests.swift

@@ -32,4 +32,24 @@ struct TalkModeConfigParsingTests {
         #expect(selection?.config["voiceId"]?.stringValue == "voice-legacy")
         #expect(selection?.config["apiKey"]?.stringValue == "legacy-key")
     }
+
+    @Test func readsConfiguredSilenceTimeoutMs() {
+        let talk: [String: AnyCodable] = [
+            "silenceTimeoutMs": AnyCodable(1500),
+        ]
+
+        #expect(TalkModeRuntime.resolvedSilenceTimeoutMs(talk) == 1500)
+    }
+
+    @Test func defaultsSilenceTimeoutMsWhenMissing() {
+        #expect(TalkModeRuntime.resolvedSilenceTimeoutMs(nil) == 700)
+    }
+
+    @Test func defaultsSilenceTimeoutMsWhenInvalid() {
+        let talk: [String: AnyCodable] = [
+            "silenceTimeoutMs": AnyCodable(0),
+        ]
+
+        #expect(TalkModeRuntime.resolvedSilenceTimeoutMs(talk) == 700)
+    }
 }

docs/gateway/configuration-reference.md

@@ -1659,6 +1659,7 @@ Defaults for Talk mode (macOS/iOS/Android).
     modelId: "eleven_v3",
     outputFormat: "mp3_44100_128",
     apiKey: "elevenlabs_api_key",
+    silenceTimeoutMs: 1500,
     interruptOnSpeech: true,
   },
 }
@@ -1668,6 +1669,7 @@ Defaults for Talk mode (macOS/iOS/Android).
 - `apiKey` and `providers.*.apiKey` accept plaintext strings or SecretRef objects.
 - `ELEVENLABS_API_KEY` fallback applies only when no Talk API key is configured.
 - `voiceAliases` lets Talk directives use friendly names.
+- `silenceTimeoutMs` controls how long Talk mode waits after user silence before it sends the transcript. Unset keeps the platform default pause window (`700` ms on macOS and Android, `900` ms on iOS).
 
 ---
 

docs/nodes/talk.md

@@ -56,6 +56,7 @@ Supported keys:
     modelId: "eleven_v3",
     outputFormat: "mp3_44100_128",
     apiKey: "elevenlabs_api_key",
+    silenceTimeoutMs: 1500,
     interruptOnSpeech: true,
   },
 }
@@ -64,6 +65,7 @@ Supported keys:
 Defaults:
 
 - `interruptOnSpeech`: true
+- `silenceTimeoutMs`: when unset, Talk keeps the platform default pause window before sending the transcript (`700` ms on macOS and Android, `900` ms on iOS)
 - `voiceId`: falls back to `ELEVENLABS_VOICE_ID` / `SAG_VOICE_ID` (or first ElevenLabs voice when API key is available)
 - `modelId`: defaults to `eleven_v3` when unset
 - `apiKey`: falls back to `ELEVENLABS_API_KEY` (or gateway shell profile if available)

src/config/schema.help.quality.test.ts

@@ -305,6 +305,7 @@ const TARGET_KEYS = [
   "talk.modelId",
   "talk.outputFormat",
   "talk.interruptOnSpeech",
+  "talk.silenceTimeoutMs",
   "meta",
   "env",
   "env.shellEnv",