Stream TTS audio to caller in real-time chunks

TTS audio was sent as a single huge WebSocket frame that overflowed the
browser's 3s ring buffer. Now streams in 60ms chunks at real-time rate.
Also increased browser ring buffer from 3s to 10s as safety net.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
2026-02-05 16:56:22 -07:00
parent 97d37f3381
commit d4e25ceb88
4 changed files with 37 additions and 6 deletions

View File

@@ -673,11 +673,11 @@ async def text_to_speech(request: TTSRequest):
)
thread.start()
# Also send to active real callers so they hear the AI
# Also stream to active real callers so they hear the AI
if session.active_real_caller:
caller_id = session.active_real_caller["caller_id"]
asyncio.create_task(
caller_service.send_audio_to_caller(caller_id, audio_bytes, 24000)
caller_service.stream_audio_to_caller(caller_id, audio_bytes, 24000)
)
return {"status": "playing", "duration": len(audio_bytes) / 2 / 24000}

View File

@@ -3,6 +3,7 @@
import asyncio
import time
import threading
import numpy as np
from typing import Optional
@@ -111,14 +112,16 @@ class CallerService:
self._websockets.pop(caller_id, None)
async def send_audio_to_caller(self, caller_id: str, pcm_data: bytes, sample_rate: int):
"""Send audio to real caller via WebSocket binary frame"""
"""Send small audio chunk to real caller via WebSocket binary frame.
For short chunks (host mic, ≤960 samples), sends immediately.
For large chunks (TTS), use stream_audio_to_caller instead.
"""
ws = self._websockets.get(caller_id)
if not ws:
return
try:
if sample_rate != 16000:
import numpy as np
audio = np.frombuffer(pcm_data, dtype=np.int16).astype(np.float32) / 32768.0
ratio = 16000 / sample_rate
out_len = int(len(audio) * ratio)
@@ -130,6 +133,34 @@ class CallerService:
except Exception as e:
print(f"[Caller] Failed to send audio: {e}")
async def stream_audio_to_caller(self, caller_id: str, pcm_data: bytes, sample_rate: int):
"""Stream large audio (TTS) to caller in real-time chunks to avoid buffer overflow."""
ws = self._websockets.get(caller_id)
if not ws:
return
try:
audio = np.frombuffer(pcm_data, dtype=np.int16).astype(np.float32) / 32768.0
if sample_rate != 16000:
ratio = 16000 / sample_rate
out_len = int(len(audio) * ratio)
indices = (np.arange(out_len) / ratio).astype(int)
indices = np.clip(indices, 0, len(audio) - 1)
audio = audio[indices]
# Send in 60ms chunks at real-time rate
chunk_samples = 960
for i in range(0, len(audio), chunk_samples):
if caller_id not in self._websockets:
break
chunk = audio[i:i + chunk_samples]
pcm_chunk = (chunk * 32767).astype(np.int16).tobytes()
await ws.send_bytes(pcm_chunk)
await asyncio.sleep(0.055) # ~60ms, slightly under to stay ahead
except Exception as e:
print(f"[Caller] Failed to stream audio: {e}")
async def notify_caller(self, caller_id: str, message: dict):
"""Send JSON control message to caller"""
ws = self._websockets.get(caller_id)

View File

@@ -150,6 +150,6 @@
</div>
</div>
<script src="/js/call-in.js?v=4"></script>
<script src="/js/call-in.js?v=5"></script>
</body>
</html>

View File

@@ -78,7 +78,7 @@ registerProcessor('caller-processor', CallerProcessor);
class PlaybackProcessor extends AudioWorkletProcessor {
constructor() {
super();
this.ringSize = 16000 * 3; // 3s ring buffer at 16kHz
this.ringSize = 16000 * 10; // 10s ring buffer at 16kHz
this.ring = new Float32Array(this.ringSize);
this.writePos = 0;
this.readPos = 0;