From 023c2ee40dc59c1474583e2dbd9f17eed9e92600 Mon Sep 17 00:00:00 2001
From: Mikei386 <44135113+Mikei386@users.noreply.github.com>
Date: Thu, 3 Sep 2026 23:19:51 +0200
Subject: [PATCH] Smooth Talk playback and gate audio tail

---
 .../openclaw-athena-talk/dist/index.js        |  8 ++++++--
 integrations/openclaw-athena-talk/index.ts    | 20 +++++++++++++------
 .../openclaw-athena-talk/openclaw.plugin.json |  2 +-
 .../openclaw-athena-talk/package.json         |  2 +-
 4 files changed, 22 insertions(+), 10 deletions(-)

diff --git a/integrations/openclaw-athena-talk/dist/index.js b/integrations/openclaw-athena-talk/dist/index.js
index 08735c5..5c527a0 100644
--- a/integrations/openclaw-athena-talk/dist/index.js
+++ b/integrations/openclaw-athena-talk/dist/index.js
@@ -237,14 +237,18 @@ class AthenaTalkBridge {
             const pcm = await this.synthesize(text, controller.signal);
             if (generation !== this.generation || !this.isConnected())
                 return;
-            const frameBytes = 24000 * 2 / 10;
+            const frameBytes = 24000 * 2 / 50;
+            const playbackStartedAt = Date.now();
             for (let offset = 0; offset < pcm.length; offset += frameBytes) {
                 if (controller.signal.aborted || generation !== this.generation || !this.isConnected()) {
                     return;
                 }
                 this.req.onAudio?.(pcm.subarray(offset, Math.min(offset + frameBytes, pcm.length)));
-                await new Promise((resolve) => setTimeout(resolve, 80));
+                await new Promise((resolve) => setTimeout(resolve, 18));
             }
+            const audioDurationMs = pcm.length / (24000 * 2) * 1000;
+            const remainingPlaybackMs = Math.max(0, audioDurationMs - (Date.now() - playbackStartedAt));
+            await new Promise((resolve) => setTimeout(resolve, remainingPlaybackMs + 300));
             this.req.onResponseDone?.({ status: "completed", responseId });
             this.req.onEvent?.({ direction: "server", type: "response.done", responseId });
         }
diff --git a/integrations/openclaw-athena-talk/index.ts b/integrations/openclaw-athena-talk/index.ts
index 7b631e2..986e98a 100644
--- a/integrations/openclaw-athena-talk/index.ts
+++ b/integrations/openclaw-athena-talk/index.ts
@@ -246,18 +246,26 @@ class AthenaTalkBridge {
       this.req.onTranscript?.("assistant", text, true);
       const pcm = await this.synthesize(text, controller.signal);
       if (generation !== this.generation || !this.isConnected()) return;
-      // Feed the relay 100 ms batches slightly faster than realtime. OpenClaw
-      // splits these into its native 20 ms frames. The small lead gives the
-      // client a jitter buffer and avoids the crackling caused by timer and
-      // WebSocket scheduling gaps at exact 20 ms pacing.
-      const frameBytes = 24000 * 2 / 10;
+      // Feed OpenClaw in its native 20 ms frame size. Sending larger bursts
+      // made the relay split several frames at once; slow-client protection
+      // could then drop individual frames and produce audible holes.
+      const frameBytes = 24000 * 2 / 50;
+      const playbackStartedAt = Date.now();
       for (let offset = 0; offset < pcm.length; offset += frameBytes) {
         if (controller.signal.aborted || generation !== this.generation || !this.isConnected()) {
           return;
         }
         this.req.onAudio?.(pcm.subarray(offset, Math.min(offset + frameBytes, pcm.length)));
-        await new Promise((resolve) => setTimeout(resolve, 80));
+        // Run just ahead of realtime so the client builds a small jitter
+        // buffer without flooding the WebSocket queue.
+        await new Promise((resolve) => setTimeout(resolve, 18));
       }
+      // Transmission is intentionally a little faster than playback. Keep the
+      // microphone gated until the client has consumed that buffered tail;
+      // otherwise the last spoken words are transcribed again as user input.
+      const audioDurationMs = pcm.length / (24000 * 2) * 1000;
+      const remainingPlaybackMs = Math.max(0, audioDurationMs - (Date.now() - playbackStartedAt));
+      await new Promise((resolve) => setTimeout(resolve, remainingPlaybackMs + 300));
       this.req.onResponseDone?.({ status: "completed", responseId });
       this.req.onEvent?.({ direction: "server", type: "response.done", responseId });
     } catch (error) {
diff --git a/integrations/openclaw-athena-talk/openclaw.plugin.json b/integrations/openclaw-athena-talk/openclaw.plugin.json
index 48e1f91..eb19c70 100644
--- a/integrations/openclaw-athena-talk/openclaw.plugin.json
+++ b/integrations/openclaw-athena-talk/openclaw.plugin.json
@@ -2,7 +2,7 @@
   "id": "athena-talk",
   "name": "Athena Local Talk",
   "description": "Local German voice conversations through Athena without a public speech provider.",
-  "version": "0.1.2",
+  "version": "0.1.3",
   "enabledByDefault": true,
   "activation": {
     "onStartup": true,
diff --git a/integrations/openclaw-athena-talk/package.json b/integrations/openclaw-athena-talk/package.json
index e713aed..b51d729 100644
--- a/integrations/openclaw-athena-talk/package.json
+++ b/integrations/openclaw-athena-talk/package.json
@@ -1,6 +1,6 @@
 {
   "name": "openclaw-plugin-athena-talk",
-  "version": "0.1.2",
+  "version": "0.1.3",
   "description": "Private realtime Talk bridge for Athena STT, OpenClaw agent consult, and Athena TTS.",
   "type": "module",
   "private": true,