talk: add configurable silence timeout

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",

src/config/schema.help.ts

@@ -163,6 +163,8 @@ export const FIELD_HELP: Record<string, string> = {
     "Use this legacy ElevenLabs API key for Talk mode only during migration, and keep secrets in env-backed storage. Prefer talk.providers.elevenlabs.apiKey (fallback: ELEVENLABS_API_KEY).",
   "talk.interruptOnSpeech":
     "If true (default), stop assistant speech when the user starts speaking in Talk mode. Keep enabled for conversational turn-taking.",
+  "talk.silenceTimeoutMs":
+    "Milliseconds of user silence before Talk mode finalizes and sends the current transcript. Leave unset to keep the platform default pause window (700 ms on macOS and Android, 900 ms on iOS).",
   acp: "ACP runtime controls for enabling dispatch, selecting backends, constraining allowed agent targets, and tuning streamed turn projection behavior.",
   "acp.enabled":
     "Global ACP feature gate. Keep disabled unless ACP runtime + policy are configured.",

src/config/schema.labels.ts

@@ -651,6 +651,7 @@ export const FIELD_LABELS: Record<string, string> = {
   "talk.modelId": "Talk Model ID",
   "talk.outputFormat": "Talk Output Format",
   "talk.interruptOnSpeech": "Talk Interrupt on Speech",
+  "talk.silenceTimeoutMs": "Talk Silence Timeout (ms)",
   messages: "Messages",
   "messages.messagePrefix": "Inbound Message Prefix",
   "messages.responsePrefix": "Outbound Response Prefix",

src/config/talk.normalize.test.ts

@@ -32,6 +32,7 @@ describe("talk normalization", () => {
       outputFormat: "pcm_44100",
       apiKey: "secret-key", // pragma: allowlist secret
       interruptOnSpeech: false,
+      silenceTimeoutMs: 1500,
     });
 
     expect(normalized).toEqual({
@@ -51,6 +52,7 @@ describe("talk normalization", () => {
       outputFormat: "pcm_44100",
       apiKey: "secret-key", // pragma: allowlist secret
       interruptOnSpeech: false,
+      silenceTimeoutMs: 1500,
     });
   });
 

src/config/talk.ts

@@ -47,6 +47,13 @@ function normalizeTalkSecretInput(value: unknown): TalkProviderConfig["apiKey"]
   return coerceSecretRef(value) ?? undefined;
 }
 
+function normalizeSilenceTimeoutMs(value: unknown): number | undefined {
+  if (typeof value !== "number" || !Number.isInteger(value) || value <= 0) {
+    return undefined;
+  }
+  return value;
+}
+
 function normalizeTalkProviderConfig(value: unknown): TalkProviderConfig | undefined {
   if (!isPlainObject(value)) {
     return undefined;
@@ -125,6 +132,10 @@ function normalizedLegacyTalkFields(source: Record<string, unknown>): Partial<Ta
   if (apiKey !== undefined) {
     legacy.apiKey = apiKey;
   }
+  const silenceTimeoutMs = normalizeSilenceTimeoutMs(source.silenceTimeoutMs);
+  if (silenceTimeoutMs !== undefined) {
+    legacy.silenceTimeoutMs = silenceTimeoutMs;
+  }
   return legacy;
 }
 
@@ -267,6 +278,9 @@ export function buildTalkConfigResponse(value: unknown): TalkConfig | undefined
   if (typeof normalized.interruptOnSpeech === "boolean") {
     payload.interruptOnSpeech = normalized.interruptOnSpeech;
   }
+  if (typeof normalized.silenceTimeoutMs === "number") {
+    payload.silenceTimeoutMs = normalized.silenceTimeoutMs;
+  }
   if (normalized.providers && Object.keys(normalized.providers).length > 0) {
     payload.providers = normalized.providers;
   }

src/config/types.gateway.ts

@@ -70,6 +70,8 @@ export type TalkConfig = {
   providers?: Record<string, TalkProviderConfig>;
   /** Stop speaking when user starts talking (default: true). */
   interruptOnSpeech?: boolean;
+  /** Milliseconds of user silence before Talk mode sends the transcript after a pause. */
+  silenceTimeoutMs?: number;
 
   /**
    * Legacy ElevenLabs compatibility fields.

src/config/zod-schema.ts

@@ -595,6 +595,7 @@ export const OpenClawSchema = z
         outputFormat: z.string().optional(),
         apiKey: SecretInputSchema.optional().register(sensitive),
         interruptOnSpeech: z.boolean().optional(),
+        silenceTimeoutMs: z.number().int().positive().optional(),
       })
       .strict()
       .optional(),

src/gateway/protocol/schema/channels.ts

@@ -42,6 +42,7 @@ export const TalkConfigResultSchema = Type.Object(
               outputFormat: Type.Optional(Type.String()),
               apiKey: Type.Optional(Type.String()),
               interruptOnSpeech: Type.Optional(Type.Boolean()),
+              silenceTimeoutMs: Type.Optional(Type.Integer({ minimum: 1 })),
             },
             { additionalProperties: false },
           ),

src/gateway/server.talk-config.test.ts

@@ -56,7 +56,11 @@ async function connectOperator(ws: GatewaySocket, scopes: string[]) {
   });
 }
 
-async function writeTalkConfig(config: { apiKey?: string; voiceId?: string }) {
+async function writeTalkConfig(config: {
+  apiKey?: string;
+  voiceId?: string;
+  silenceTimeoutMs?: number;
+}) {
   const { writeConfigFile } = await import("../config/config.js");
   await writeConfigFile({ talk: config });
 }
@@ -68,6 +72,7 @@ describe("gateway talk.config", () => {
       talk: {
         voiceId: "voice-123",
         apiKey: "secret-key-abc", // pragma: allowlist secret
+        silenceTimeoutMs: 1500,
       },
       session: {
         mainKey: "main-test",
@@ -88,6 +93,7 @@ describe("gateway talk.config", () => {
             };
             apiKey?: string;
             voiceId?: string;
+            silenceTimeoutMs?: number;
           };
         };
       }>(ws, "talk.config", {});
@@ -99,6 +105,7 @@ describe("gateway talk.config", () => {
       );
       expect(res.payload?.config?.talk?.voiceId).toBe("voice-123");
       expect(res.payload?.config?.talk?.apiKey).toBe("__OPENCLAW_REDACTED__");
+      expect(res.payload?.config?.talk?.silenceTimeoutMs).toBe(1500);
     });
   });