From 0f9b43c8ba6506aa2f0a3cc12abfb07c2dad5fb3 Mon Sep 17 00:00:00 2001
From: jenkins <brad@bstein.dev>
Date: Sun, 23 Aug 2026 16:50:48 -0300
Subject: [PATCH] perf(hermes-voice): reduce conversational latency

---
 dockerfiles/hermes-webui-atlas-voice.js   | 11 ++++++++---
 testing/tests/test_hermes_chat_quality.py |  4 +++-
 2 files changed, 11 insertions(+), 4 deletions(-)

diff --git a/dockerfiles/hermes-webui-atlas-voice.js b/dockerfiles/hermes-webui-atlas-voice.js
index f15a0ef4..b8f0aadd 100644
--- a/dockerfiles/hermes-webui-atlas-voice.js
+++ b/dockerfiles/hermes-webui-atlas-voice.js
@@ -252,7 +252,7 @@
       // intermittently unreadable by ffmpeg. A bounded 90-second Opus capture
       // is small enough to retain as one browser-owned recording.
       recorder.start();
-      const silenceMs=Math.max(900,parseInt(localStorage.getItem('hermes-voice-silence-ms')||'1600',10)||1600);
+      const silenceMs=Math.max(900,parseInt(localStorage.getItem('hermes-voice-silence-ms')||'1100',10)||1100);
       vadTimer=window.setInterval(function(){
         if(!active||token!==generation||!recorder||recorder.state==='inactive') return;
         analyser.getByteTimeDomainData(samples);
@@ -349,7 +349,7 @@
     setState('speaking');
     // Shorter first chunks lower time-to-first-audio; the next chunk is
     // synthesized while the current one plays.
-    const chunks=typeof window._splitForTTS==='function'?window._splitForTTS(text,160):[text];
+    const chunks=typeof window._splitForTTS==='function'?window._splitForTTS(text,80):[text];
     try{
       let pending=fetchSpeech(chunks[0],language);
       for(let index=0;index<chunks.length;index+=1){
@@ -399,7 +399,12 @@
         localStorage.setItem('hermes-voice-mode-button','true');
         localStorage.setItem('hermes-tts-engine','atlas');
         localStorage.setItem('hermes-tts-enabled','true');
-        localStorage.setItem('hermes-voice-silence-ms','1600');
+      }
+      // Move existing installations to the lower-latency silence window once;
+      // subsequent user changes remain untouched.
+      if(localStorage.getItem('hermes-atlas-voice-latency-v2')!=='1'){
+        localStorage.setItem('hermes-atlas-voice-latency-v2','1');
+        localStorage.setItem('hermes-voice-silence-ms','1100');
       }
       const selector=document.getElementById('settingsTtsEngine');
       if(selector&&!selector.querySelector('option[value="atlas"]')){
diff --git a/testing/tests/test_hermes_chat_quality.py b/testing/tests/test_hermes_chat_quality.py
index 03cd76ef..d6fdca7e 100644
--- a/testing/tests/test_hermes_chat_quality.py
+++ b/testing/tests/test_hermes_chat_quality.py
@@ -342,7 +342,9 @@ def test_chat_voice_uses_private_jetson_services_and_shared_auto_route():
     assert "/api/transcribe" in voice_script
     assert "/api/tts" in voice_script
     assert "speakResponse(generation)" in voice_script
-    assert "window._splitForTTS(text,160)" in voice_script
+    assert "window._splitForTTS(text,80)" in voice_script
+    assert "hermes-atlas-voice-latency-v2" in voice_script
+    assert "hermes-voice-silence-ms','1100'" in voice_script
     assert "pending=fetchSpeech(chunks[index+1],language)" in voice_script
     assert "restartSoon(token,450)" in voice_script
     assert "constraints.voiceIsolation=true" in voice_script