Fix voice-assistant agent routing and TTS escape sequence handling

- Resolve agent route via bindings (matching Telegram/Discord pattern) so voice-agent binding in openclaw.json is respected
- Handle literal escape sequences (\n, \r, \t) before stripping backslashes in TTS sanitizer

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: sanchitmonga22

extensions/voice-assistant/src/channel.ts

@@ -18,8 +18,8 @@ import type {
   VoiceAssistantConfig,
   VoiceSpeakMessage,
 } from "./types.js";
-import { getVoiceAssistantRuntime } from "./runtime.js";
 import { createVoiceGateway, getVoiceGateway, stopVoiceGateway } from "./gateway.js";
+import { getVoiceAssistantRuntime } from "./runtime.js";
 import { sanitizeForTTS } from "./tts-sanitize.js";
 
 // =============================================================================
@@ -131,79 +131,95 @@ export const voiceAssistantPlugin: ChannelPlugin<ResolvedVoiceAssistantAccount>
       const gateway = createVoiceGateway(wsPort);
 
       // Handle transcriptions from voice clients
-      gateway.on("transcription", async (event: {
-        clientId: string;
-        accountId: string;
-        text: string;
-        sessionId?: string;
-        timestamp: number;
-        isFinal: boolean;
-      }) => {
-        if (!event.isFinal) {
-          // Skip partial transcriptions
-          return;
-        }
+      gateway.on(
+        "transcription",
+        async (event: {
+          clientId: string;
+          accountId: string;
+          text: string;
+          sessionId?: string;
+          timestamp: number;
+          isFinal: boolean;
+        }) => {
+          if (!event.isFinal) {
+            // Skip partial transcriptions
+            return;
+          }
 
-        log?.info?.(`Transcription received: "${event.text}"`);
-
-        try {
-          const sessionKey = `voice:${event.accountId}:${event.sessionId ?? "main"}`;
-          const from = `voice:${event.clientId}`;
-          const agentRuntime = getVoiceAssistantRuntime();
-
-          // Load latest config for dispatch
-          const currentCfg = await agentRuntime.config.loadConfig();
-
-          // Build MsgContext matching the pattern used by Telegram/Discord
-          const msgCtx = {
-            Body: event.text,
-            BodyForAgent: event.text,
-            RawBody: event.text,
-            CommandBody: event.text,
-            From: from,
-            To: CHANNEL_ID,
-            SessionKey: sessionKey,
-            AccountId: event.accountId,
-            ChatType: "direct",
-            Provider: CHANNEL_ID,
-            Surface: CHANNEL_ID,
-            OriginatingChannel: CHANNEL_ID,
-            OriginatingTo: from,
-            SenderName: account.name || "Voice User",
-            SenderId: event.clientId,
-            Timestamp: event.timestamp,
-            WasMentioned: true, // Direct voice input is always "mentioned"
-          };
-
-          // Dispatch using the same API as Telegram/Discord
-          await agentRuntime.channel.reply.dispatchReplyWithBufferedBlockDispatcher({
-            ctx: msgCtx,
-            cfg: currentCfg,
-            dispatcherOptions: {
-              deliver: async (payload) => {
-                if (payload.text) {
-                  const spokenText = sanitizeForTTS(payload.text);
-                  if (!spokenText) return;
-                  log?.info?.(`Sending TTS response: "${spokenText.slice(0, 80)}..."`);
-                  const speakMsg: VoiceSpeakMessage = {
-                    type: "speak",
-                    text: spokenText,
-                    sourceChannel: CHANNEL_ID,
-                    priority: 1,
-                    interrupt: false,
-                  };
-                  gateway.broadcast(speakMsg);
-                }
+          log?.info?.(`Transcription received: "${event.text}"`);
+
+          try {
+            const from = `voice:${event.clientId}`;
+            const agentRuntime = getVoiceAssistantRuntime();
+
+            // Load latest config for dispatch
+            const currentCfg = await agentRuntime.config.loadConfig();
+
+            // Resolve agent route via bindings (same pattern as Telegram/Discord)
+            // This ensures voice-agent binding in openclaw.json is respected,
+            // routing to the correct agent workspace + SOUL.md
+            const route = agentRuntime.channel.routing.resolveAgentRoute({
+              cfg: currentCfg,
+              channel: CHANNEL_ID,
+              accountId: event.accountId,
+              peer: {
+                kind: "direct" as "direct",
+                id: from,
               },
-              onError: (err) => {
-                log?.error?.(`Voice reply delivery error: ${err}`);
+            });
+            const sessionKey = route.sessionKey;
+
+            // Build MsgContext matching the pattern used by Telegram/Discord
+            const msgCtx = {
+              Body: event.text,
+              BodyForAgent: event.text,
+              RawBody: event.text,
+              CommandBody: event.text,
+              From: from,
+              To: CHANNEL_ID,
+              SessionKey: sessionKey,
+              AccountId: route.accountId,
+              ChatType: "direct",
+              Provider: CHANNEL_ID,
+              Surface: CHANNEL_ID,
+              OriginatingChannel: CHANNEL_ID,
+              OriginatingTo: from,
+              SenderName: account.name || "Voice User",
+              SenderId: event.clientId,
+              Timestamp: event.timestamp,
+              WasMentioned: true, // Direct voice input is always "mentioned"
+            };
+
+            // Dispatch using the same API as Telegram/Discord
+            await agentRuntime.channel.reply.dispatchReplyWithBufferedBlockDispatcher({
+              ctx: msgCtx,
+              cfg: currentCfg,
+              dispatcherOptions: {
+                deliver: async (payload) => {
+                  if (payload.text) {
+                    const spokenText = sanitizeForTTS(payload.text);
+                    if (!spokenText) return;
+                    log?.info?.(`Sending TTS response: "${spokenText.slice(0, 80)}..."`);
+                    const speakMsg: VoiceSpeakMessage = {
+                      type: "speak",
+                      text: spokenText,
+                      sourceChannel: CHANNEL_ID,
+                      priority: 1,
+                      interrupt: false,
+                    };
+                    gateway.broadcast(speakMsg);
+                  }
+                },
+                onError: (err) => {
+                  log?.error?.(`Voice reply delivery error: ${err}`);
+                },
               },
-            },
-          });
-        } catch (error) {
-          log?.error?.(`Failed to dispatch transcription: ${error}`);
-        }
-      });
+            });
+          } catch (error) {
+            log?.error?.(`Failed to dispatch transcription: ${error}`);
+          }
+        },
+      );
 
       // Handle abort signal
       abortSignal.addEventListener("abort", () => {

extensions/voice-assistant/src/tts-sanitize.ts

@@ -76,6 +76,11 @@ export function sanitizeForTTS(text: string): string {
 
   // --- Clean up special characters ---
 
+  // Literal escape sequences (\n, \r, \t) → actual characters before stripping backslashes
+  result = result.replace(/\\n/g, "\n");
+  result = result.replace(/\\r/g, "");
+  result = result.replace(/\\t/g, " ");
+
   // HTML-like tags (angle brackets)
   result = result.replace(/<[^>]*>/g, "");