Telephony Integration (Twilio, etc.)

Updated November 3, 2026 based on the Realtime API documentation: the G.711 μ-law format is named audio/pcmu (the text previously cited g711_ulaw, which was the beta name). Source: Realtime API.

November 2, 2026 update — Audio models. Several models in this course are deprecated: tts-1, tts-1-hd, and gpt-4o-mini-tts will stop working on January 6, 2027 (the indicated replacement is gpt-realtime-2.1-mini, which the OpenAI TTS guide states is not supported by /v1/audio/speech: speech synthesis migration must go through the Realtime API); gpt-realtime, gpt-4o-realtime, and legacy audio models on January 20, 2027 (gpt-realtime-2.1, gpt-realtime-2.1-mini, gpt-audio-1.5); whisper-1, gpt-4o-transcribe, and /v1/audio/translations on February 26, 2027 (gpt-transcribe, gpt-live-transcribe, v1/realtime/translations). For full-duplex voice, GPT-Live 1 has been available in the API since September 10, 2026.

Telephony Integration (Twilio, etc.)

A voice agent becomes truly useful when it’s accessible by phone. Cloud telephony platforms like Twilio, LiveKit, and Daily let you connect the Realtime API to the global telephone network. A user dials a number, and your agent answers in real time.

Telephony Architecture

Telephony integration adds a layer between the telephone network and the Realtime API:

Implementation with Twilio

Incoming Call Webhook

When a call arrives, Twilio sends an HTTP request to your server. You respond with TwiML that opens a Media Stream:

from fastapi import FastAPI, Request
from fastapi.responses import Response

app = FastAPI()

@app.post("/incoming-call")
async def incoming_call(request: Request):
    """Twilio webhook for incoming calls."""
    twiml = """<?xml version="1.0" encoding="UTF-8"?>
    <Response>
        <Say language="en-US">
            Welcome to Corsen AI. Connecting you to your voice assistant.
        </Say>
        <Connect>
            <Stream url="wss://your-server.com/media-stream">
                <Parameter name="caller" value="{caller_number}" />
            </Stream>
        </Connect>
    </Response>"""

    form_data = await request.form()
    caller = form_data.get("From", "unknown")
    twiml = twiml.replace("{caller_number}", caller)

    return Response(content=twiml, media_type="application/xml")

The Media Stream ↔ Realtime API Bridge

The WebSocket server receives audio from Twilio and relays it to the Realtime API:

from fastapi import WebSocket
import websockets
import json
import base64
import audioop
import asyncio

OPENAI_API_KEY = "sk-..."
REALTIME_URL = "wss://api.openai.com/v1/realtime?model=gpt-realtime-2.1"

@app.websocket("/media-stream")
async def media_stream(twilio_ws: WebSocket):
    """Bridge between Twilio Media Stream and the Realtime API."""
    await twilio_ws.accept()

    # Connect to the Realtime API
    headers = {
        "Authorization": f"Bearer {OPENAI_API_KEY}"
    }

    async with websockets.connect(REALTIME_URL, additional_headers=headers) as openai_ws:
        await openai_ws.recv()  # session.created

        # Configure the session
        await openai_ws.send(json.dumps({
            "type": "session.update",
            "session": {
                "type": "realtime",
                "output_modalities": ["audio"],
                "instructions": "You are a phone assistant. "
                                "Respond in English, concisely.",
                "audio": {
                    "input": {
                        "format": {"type": "audio/pcmu"},
                        "turn_detection": {
                            "type": "server_vad",
                            "threshold": 0.5,
                            "silence_duration_ms": 600
                        }
                    },
                    "output": {
                        "format": {"type": "audio/pcmu"},
                        "voice": "marin"
                    }
                }
            }
        }))
        await openai_ws.recv()

        stream_sid = None

        async def twilio_to_openai():
            nonlocal stream_sid
            async for message in twilio_ws.iter_text():
                data = json.loads(message)

                if data["event"] == "media":
                    audio_b64 = data["media"]["payload"]
                    await openai_ws.send(json.dumps({
                        "type": "input_audio_buffer.append",
                        "audio": audio_b64
                    }))

                elif data["event"] == "start":
                    stream_sid = data["start"]["streamSid"]

        async def openai_to_twilio():
            async for message in openai_ws:
                event = json.loads(message)

                if event["type"] == "response.output_audio.delta":
                    await twilio_ws.send_json({
                        "event": "media",
                        "streamSid": stream_sid,
                        "media": {
                            "payload": event["delta"]
                        }
                    })

        await asyncio.gather(twilio_to_openai(), openai_to_twilio())

G.711 Audio Format

Telephony uses the G.711 μ-law format (8 kHz, one byte per sample). The Realtime API supports this natively under the name audio/pcmu, which you declare for both input and output (audio.input.format and audio.output.format, as in the code above). This avoids any server-side conversion. The name g711_ulaw belonged to the API beta; the A-law equivalent is named audio/pcma.

Twilio Alternatives

Key Takeaways

• Twilio Media Streams lets you connect a phone call to a bidirectional WebSocket
• Your server acts as a bridge between the Twilio Media Stream and the OpenAI Realtime API
• The G.711 μ-law format is native to telephony and directly supported by the Realtime API
• LiveKit (open source, WebRTC) and Daily (hosted) are viable alternatives to Twilio
• The architecture is always the same: telephony → your server → Realtime API