Fix choppy/distorted audio to live caller

- Mute host mic forwarding while TTS is streaming to prevent interleaving
  both audio sources into the same playback buffer
- Replace nearest-neighbor downsampling with box-filter averaging on both
  server (host mic) and browser (caller mic) for anti-aliased resampling

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

backend/main.py

@@ -880,6 +880,9 @@ async def _host_audio_sender():
     """Persistent task that drains audio queue and sends to callers"""
     while True:
         pcm_bytes = await _host_audio_queue.get()
+        # Skip host mic audio while TTS is streaming to avoid interleaving
+        if caller_service.streaming_tts:
+            continue
         for caller_id in list(caller_service.active_calls.keys()):
             try:
                 await caller_service.send_audio_to_caller(caller_id, pcm_bytes, 16000)

backend/services/audio.py

@@ -472,9 +472,10 @@ class AudioService:
                 if not self._host_send_callback:
                     return
                 mono = indata[:, record_channel]
-                # Simple decimation to ~16kHz
+                # Downsample to ~16kHz with averaging (anti-aliased)
                 if step > 1:
-                    mono = mono[::step]
+                    n = len(mono) // step * step
+                    mono = mono[:n].reshape(-1, step).mean(axis=1)
 
                 host_accum.append(mono.copy())
                 host_accum_samples[0] += len(mono)

backend/services/caller_service.py

@@ -19,6 +19,7 @@ class CallerService:
         self._caller_counter: int = 0
         self._lock = threading.Lock()
         self._websockets: dict[str, any] = {}  # caller_id -> WebSocket
+        self.streaming_tts: bool = False  # True while TTS audio is being streamed
 
     def add_to_queue(self, caller_id: str, name: str):
         with self._lock:
@@ -139,6 +140,7 @@ class CallerService:
         if not ws:
             return
 
+        self.streaming_tts = True
         try:
             audio = np.frombuffer(pcm_data, dtype=np.int16).astype(np.float32) / 32768.0
             if sample_rate != 16000:
@@ -160,6 +162,8 @@ class CallerService:
 
         except Exception as e:
             print(f"[Caller] Failed to stream audio: {e}")
+        finally:
+            self.streaming_tts = False
 
     async def notify_caller(self, caller_id: str, message: dict):
         """Send JSON control message to caller"""

frontend/call-in.html

@@ -150,6 +150,6 @@
         </div>
     </div>
 
-    <script src="/js/call-in.js?v=5"></script>
+    <script src="/js/call-in.js?v=6"></script>
 </body>
 </html>

frontend/js/call-in.js

@@ -52,12 +52,14 @@ class CallerProcessor extends AudioWorkletProcessor {
         const input = inputs[0][0];
         if (!input) return true;
 
-        const ratio = sampleRate / 16000;
+        // Downsample with averaging (anti-aliased)
-        for (let i = 0; i < input.length; i += ratio) {
+        const step = Math.floor(sampleRate / 16000);
-            const idx = Math.floor(i);
+        for (let i = 0; i + step <= input.length; i += step) {
-            if (idx < input.length) {
+            let sum = 0;
-                this.buffer.push(input[idx]);
+            for (let j = 0; j < step; j++) {
+                sum += input[i + j];
             }
+            this.buffer.push(sum / step);
         }
 
         while (this.buffer.length >= this.targetSamples) {