feat: VoiceCoach v2.1 - GPU local, Ollama LLM, settings panel, token server separation

- pipeline.py: switched LLM from OpenClaw/Claude to local Ollama, added settings loading, sentence aggregator, async SpeechBrain, tuned VAD
- main.py: simplified (no more token server thread, just pipeline)
- token_server.py: NEW - standalone token + settings API server
- docker-compose.yml: all services on GPU machine, added token-server service, CUDA speaches, Ollama LLM
- web-client/index.html: added settings panel with sliders
- web-client/app.js: settings management, localStorage persistence, push to server
- web-client/style.css: settings panel styles
- web-client/nginx.conf: proxy to token-server instead of pipecat-agent
- patch_livekit.py: updated with proper indentation-safe patching

web-client/index.html

@@ -30,6 +30,60 @@
       <button class="mode-btn" onclick="setMode('practice mode')">Practice</button>
       <button class="mode-btn" onclick="setMode('briefing mode')">Briefing</button>
     </div>
+
+    <div class="settings-panel">
+      <div class="settings-header" onclick="toggleSettings()">
+        <span>Settings</span>
+        <span id="settingsToggle">+</span>
+      </div>
+      <div id="settingsBody" class="settings-body" style="display:none">
+        <div class="setting-group">
+          <label>VAD Confidence <span class="setting-val" id="val-confidence">0.7</span></label>
+          <input type="range" id="set-confidence" min="0.3" max="0.95" step="0.05" value="0.7">
+          <div class="setting-hint">Higher = less false triggers from noise</div>
+        </div>
+        <div class="setting-group">
+          <label>Min Volume <span class="setting-val" id="val-min_volume">0.3</span></label>
+          <input type="range" id="set-min_volume" min="0.05" max="0.6" step="0.05" value="0.3">
+          <div class="setting-hint">Higher = ignores quieter sounds</div>
+        </div>
+        <div class="setting-group">
+          <label>Start Delay <span class="setting-val" id="val-start_secs">0.4</span>s</label>
+          <input type="range" id="set-start_secs" min="0.1" max="0.8" step="0.1" value="0.4">
+          <div class="setting-hint">Sustained speech needed before trigger</div>
+        </div>
+        <div class="setting-group">
+          <label>Stop Delay <span class="setting-val" id="val-stop_secs">0.5</span>s</label>
+          <input type="range" id="set-stop_secs" min="0.2" max="1.5" step="0.1" value="0.5">
+          <div class="setting-hint">Silence before turn ends (higher = less cutoff)</div>
+        </div>
+        <div class="setting-group">
+          <label>LLM Max Tokens <span class="setting-val" id="val-max_tokens">150</span></label>
+          <input type="range" id="set-max_tokens" min="50" max="500" step="25" value="150">
+          <div class="setting-hint">Longer = more detailed responses, slower</div>
+        </div>
+        <div class="setting-group">
+          <label>TTS Voice <span class="setting-val" id="val-tts_voice">af_nova</span></label>
+          <select id="set-tts_voice">
+            <option value="af_nova">Nova (Female, American)</option>
+            <option value="af_sky">Sky (Female, American)</option>
+            <option value="am_adam">Adam (Male, American)</option>
+            <option value="bf_emma">Emma (Female, British)</option>
+            <option value="bm_george">George (Male, British)</option>
+          </select>
+        </div>
+        <button class="btn btn-small" onclick="saveSettings()">Apply Settings</button>
+        <button class="btn btn-small btn-reset" onclick="resetSettings()">Reset Defaults</button>
+      </div>
+    </div>
+
+    <div class="log-panel">
+      <div class="log-header">
+        <span>Log</span>
+        <button class="log-clear" onclick="clearLog()">Clear</button>
+      </div>
+      <div id="logOutput" class="log-output"></div>
+    </div>
   </div>
 
   <script src="https://cdn.jsdelivr.net/npm/livekit-client/dist/livekit-client.umd.min.js"></script>