Speech Engine quickstart

1
pip install elevenlabs openai python-dotenv

1
import asyncio
2
from dotenv import load_dotenv
3
from elevenlabs import AsyncElevenLabs
4
5
load_dotenv()
6
7
elevenlabs = AsyncElevenLabs(
8
    api_key=os.getenv("ELEVENLABS_API_KEY"),
9
)
10
11
12
async def main():
13
    engine = await elevenlabs.speech_engine.create(
14
        name="My Speech Engine",
15
        speech_engine={
16
            # Note we use the wss protocol instead of https
17
            "ws_url": "wss://abc123.ngrok.io/ws",
18
        },
19
    )
20
21
    print(f"Speech Engine ID: {engine.engine_id}")
22
23
24
if __name__ == "__main__":
25
    asyncio.run(main())

1
import asyncio
2
import os
3
4
from dotenv import load_dotenv
5
from openai import AsyncOpenAI
6
from elevenlabs import AsyncElevenLabs
7
8
load_dotenv()
9
10
# Replace with your Speech Engine ID from step 4
11
SPEECH_ENGINE_ID = "seng_8k3m9xr4hjnfg983brhmhkd98n6"
12
13
openai = AsyncOpenAI(
14
  api_key=os.getenv("OPENAI_API_KEY"),
15
)
16
elevenlabs = AsyncElevenLabs(
17
  api_key=os.getenv("ELEVENLABS_API_KEY"),
18
)
19
20
21
def on_init(conversation_id, session):
22
    print(f"Session started: {conversation_id}")
23
24
25
async def on_transcript(transcript, session):
26
    stream = await openai.responses.create(
27
        model="gpt-4o",
28
        instructions="You are a helpful voice assistant. Keep responses concise and conversational.",
29
        input=[
30
            {"role": "assistant" if m.role == "agent" else m.role, "content": m.content}
31
            for m in transcript
32
        ],
33
        stream=True,
34
    )
35
36
    await session.send_response(stream)
37
38
39
def on_close(session):
40
    print(f"Session ended: {session.conversation_id}")
41
42
43
def on_error(err, session):
44
    print(f"Error: {err}")
45
46
47
async def main():
48
    engine = await elevenlabs.speech_engine.get(SPEECH_ENGINE_ID)
49
50
    await engine.serve(
51
        port=3001,
52
        path="/ws",
53
        debug=True,
54
        on_init=on_init,
55
        on_transcript=on_transcript,
56
        on_close=on_close,
57
        on_error=on_error,
58
    )
59
60
61
if __name__ == "__main__":
62
    asyncio.run(main())

1
python server.py

1
import os
2
3
from dotenv import load_dotenv
4
from flask import Flask, jsonify
5
from elevenlabs import ElevenLabs
6
7
load_dotenv()
8
9
app = Flask(__name__)
10
elevenlabs = ElevenLabs(
11
    api_key=os.getenv("ELEVENLABS_API_KEY"),
12
)
13
14
15
@app.route("/api/token")
16
def get_token():
17
    # Replace with your Speech Engine ID from step 4 of the server setup
18
    speech_engine_id = "seng_8k3m9xr4hjnfg983brhmhkd98n6"
19
20
    response = elevenlabs.conversational_ai.conversations.get_webrtc_token(
21
        agent_id=speech_engine_id,
22
    )
23
24
    return jsonify(token=response.token)
25
26
27
if __name__ == "__main__":
28
    app.run(port=3002)

1
engine = await elevenlabs.speech_engine.update(
2
    speech_engine_id="seng_8k3m9xr4hjnfg983brhmhkd98n6",
3
    overrides={
4
      "first_message": True,
5
    },
6
)