fix(talk): prevent double TTS playback when system voice times out (#53511)

Merged via squash.

Prepared head SHA: 864d556fa6
Co-authored-by: hongsw <1100974+hongsw@users.noreply.github.com>
Co-authored-by: grp06 <1573959+grp06@users.noreply.github.com>
Reviewed-by: @grp06

apps/shared/OpenClawKit/Sources/OpenClawKit/TalkSystemSpeechSynthesizer.swift

@@ -51,11 +51,11 @@ public final class TalkSystemSpeechSynthesizer: NSObject {
         }
         self.currentUtterance = utterance
 
-        let estimatedSeconds = max(3.0, min(180.0, Double(trimmed.count) * 0.08))
+        let watchdogTimeout = Self.watchdogTimeoutSeconds(text: trimmed, language: language ?? utterance.voice?.language)
         self.watchdog?.cancel()
         self.watchdog = Task { @MainActor [weak self] in
             guard let self else { return }
-            try? await Task.sleep(nanoseconds: UInt64(estimatedSeconds * 1_000_000_000))
+            try? await Task.sleep(nanoseconds: UInt64(watchdogTimeout * 1_000_000_000))
             if Task.isCancelled { return }
             guard self.currentToken == token else { return }
             if self.synth.isSpeaking {
@@ -63,7 +63,7 @@ public final class TalkSystemSpeechSynthesizer: NSObject {
             }
             self.finishCurrent(
                 with: NSError(domain: "TalkSystemSpeechSynthesizer", code: 408, userInfo: [
-                    NSLocalizedDescriptionKey: "system TTS timed out after \(estimatedSeconds)s",
+                    NSLocalizedDescriptionKey: "system TTS timed out after \(watchdogTimeout)s",
                 ]))
         }
 
@@ -83,6 +83,37 @@ public final class TalkSystemSpeechSynthesizer: NSObject {
         }
     }
 
+    static func watchdogTimeoutSeconds(text: String, language: String?) -> Double {
+        // Estimate speech duration per language, then apply 3x safety margin.
+        // The watchdog is a hang guard — normal completion relies on didFinish.
+        //
+        // Speech rates based on Pellegrino et al. (2019) syllable-per-second data,
+        // adjusted for TTS synthesis (slower than natural speech):
+        // https://www.science.org/doi/10.1126/sciadv.aaw2594
+        //   Japanese: 7.84 SPS -> ~0.20s/char (mixed kana/kanji avg ~1.5 mora/char)
+        //   Korean:   5.96 SPS -> ~0.25s/char (1 char = 1 syllable)
+        //   Chinese:  5.18 SPS -> ~0.28s/char (1 char = 1 syllable)
+        //   English:  6.19 SPS -> ~0.08s/char (avg ~5 chars/syllable)
+        let normalizedLanguage = language?.lowercased() ?? "en"
+        let perCharSeconds: Double
+        let minSeconds: Double
+        if normalizedLanguage.hasPrefix("ko") {
+            perCharSeconds = 0.25
+            minSeconds = 10.0
+        } else if normalizedLanguage.hasPrefix("zh") {
+            perCharSeconds = 0.28
+            minSeconds = 10.0
+        } else if normalizedLanguage.hasPrefix("ja") {
+            perCharSeconds = 0.20
+            minSeconds = 10.0
+        } else {
+            perCharSeconds = 0.08
+            minSeconds = 3.0
+        }
+        let estimatedSeconds = max(minSeconds, min(300.0, Double(text.count) * perCharSeconds))
+        return estimatedSeconds * 3.0
+    }
+
     private func matchesCurrentUtterance(_ utteranceID: ObjectIdentifier) -> Bool {
         guard let currentUtterance = self.currentUtterance else { return false }
         return ObjectIdentifier(currentUtterance) == utteranceID

apps/shared/OpenClawKit/Tests/OpenClawKitTests/TalkSystemSpeechSynthesizerTests.swift

@@ -0,0 +1,44 @@
+import XCTest
+@testable import OpenClawKit
+
+final class TalkSystemSpeechSynthesizerTests: XCTestCase {
+    func testWatchdogTimeoutDefaultsToLatinProfile() {
+        let timeout = TalkSystemSpeechSynthesizer.watchdogTimeoutSeconds(
+            text: String(repeating: "a", count: 100),
+            language: nil)
+
+        XCTAssertEqual(timeout, 24.0, accuracy: 0.001)
+    }
+
+    func testWatchdogTimeoutUsesKoreanProfile() {
+        let timeout = TalkSystemSpeechSynthesizer.watchdogTimeoutSeconds(
+            text: String(repeating: "가", count: 100),
+            language: "ko-KR")
+
+        XCTAssertEqual(timeout, 75.0, accuracy: 0.001)
+    }
+
+    func testWatchdogTimeoutUsesChineseProfile() {
+        let timeout = TalkSystemSpeechSynthesizer.watchdogTimeoutSeconds(
+            text: String(repeating: "你", count: 100),
+            language: "zh-CN")
+
+        XCTAssertEqual(timeout, 84.0, accuracy: 0.001)
+    }
+
+    func testWatchdogTimeoutUsesJapaneseProfile() {
+        let timeout = TalkSystemSpeechSynthesizer.watchdogTimeoutSeconds(
+            text: String(repeating: "あ", count: 100),
+            language: "ja-JP")
+
+        XCTAssertEqual(timeout, 60.0, accuracy: 0.001)
+    }
+
+    func testWatchdogTimeoutClampsVeryLongUtterances() {
+        let timeout = TalkSystemSpeechSynthesizer.watchdogTimeoutSeconds(
+            text: String(repeating: "a", count: 10_000),
+            language: "en-US")
+
+        XCTAssertEqual(timeout, 900.0, accuracy: 0.001)
+    }
+}