Realtime Protocol

# pip install websockets
import asyncio
import json
import urllib.request
import wave

import websockets

API = "https://api.voice213.com"
AGENT_ID = "k3x9m2p7q4wz"


def call_ticket():
    request = urllib.request.Request(f"{API}/api/agents/public/{AGENT_ID}/call", method="POST")
    with urllib.request.urlopen(request, timeout=15) as response:
        return json.load(response)


async def main():
    with wave.open("question.wav", "rb") as wav:  # 16 kHz, 16-bit, mono
        question = wav.readframes(wav.getnframes())

    ticket = call_ticket()
    voice = bytearray()
    agent_said = asyncio.Event()
    ended = asyncio.Event()

    async with websockets.connect(f"{ticket['wsUrl']}?token={ticket['token']}", max_size=None) as ws:

        async def receive():
            async for message in ws:
                if isinstance(message, bytes):
                    voice.extend(message)  # the agent's voice: 24 kHz, 16-bit, mono
                    continue
                event = json.loads(message)
                if event["type"] == "ready":
                    print("connected:", event["conversationId"])
                elif event["type"] == "transcript" and event["final"]:
                    print(f"{event['role']}: {event['text']}")
                    if event["role"] == "agent":
                        agent_said.set()
                elif event["type"] == "error":
                    print("refused:", event["code"], event["error"])
                    break
                elif event["type"] == "ended":
                    print("ended:", event["reason"], f"after {event['durationSeconds']} s")
                    break
            ended.set()

        receiver = asyncio.create_task(receive())
        await asyncio.wait_for(agent_said.wait(), 30)  # its greeting (leave out if it waits for the caller)
        agent_said.clear()

        # The question in real time, 40 ms (1,280 bytes) at a time, then 2 s of silence to end the turn
        audio = question + bytes(2 * 16000 * 2)
        for at in range(0, len(audio), 1280):
            await ws.send(audio[at : at + 1280])
            await asyncio.sleep(0.04)

        await asyncio.wait_for(agent_said.wait(), 30)  # its answer
        await asyncio.sleep(2)
        await ws.send(json.dumps({"type": "end"}))
        await asyncio.wait_for(ended.wait(), 10)
        await receiver

    with wave.open("answer.wav", "wb") as out:
        out.setnchannels(1)
        out.setsampwidth(2)
        out.setframerate(24000)
        out.writeframes(voice)


asyncio.run(main())