Skip to main content
Fastest path: run /scenarios add voice testing to my agent in your coding assistant after installing the /scenarios skill — it detects your transport, picks the matching adapter, and wires it to your deployed agent. Prefer to do it by hand? Follow the steps below.

​
Testing Voice Agents

Scenario tests voice agents end-to-end over real audio — it synthesizes a caller, speaks to your agent through its real transport, and judges the conversation. It’s the same scenario.run() and the same JudgeAgent you use for text — only the medium changes. Voice Agent Testing

​
The highlights

Real-audio, multi-turn calls

A synthesized user actually speaks to your agent and reacts to its replies, turn after turn.

Audio effects

Inject background noise, phone-quality codec degradation, or custom WAV clips to test robustness.

Interruptions

Script the user barging in mid-reply — native barge-in or VAD-driven fallback per adapter.

Latency metrics

Time-to-first-byte and p50/p95 per turn, so you catch a slow agent before users do.

​
Works with your stack

One API, every major voice transport — pick the adapter that matches what you’ve already deployed:
TransportAdapter
Pipecat / Twilio Media StreamsPipecatAgentAdapter
ElevenLabs hosted ConvAIElevenLabsAgentAdapter
OpenAI RealtimeOpenAIRealtimeAgentAdapter
Gemini LiveGeminiLiveAgentAdapter
Twilio phone number (PSTN)TwilioAgentAdapter
Text-only agent (no transport yet)ComposableVoiceAgent
Full constructors, the per-adapter capability matrix, and guidance on choosing an adapter live in the Scenario docs.

​
Get it running

New to Scenario? Start with Getting Started for the basics (API key, writing your agent’s call() adapter), then come back for the voice specifics.
1

Install Scenario

pip install langwatch-scenario
2

Point at your agent, add a voice user and a judge

Swap BOT_WS_URL for your running bot (or use the adapter for your transport from the table above). The user simulator speaks with a real voice; the judge scores the call against your criteria.
import scenario

scenario.configure(default_model="openai/gpt-5-mini")

BOT_WS_URL = "ws://localhost:8765/stream"  # your deployed bot

result = await scenario.run(
    name="billing inquiry",
    description="An upset customer calls about a duplicate charge.",
    agents=[
        scenario.PipecatAgentAdapter(url=BOT_WS_URL),
        scenario.UserSimulatorAgent(
            voice="elevenlabs/EXAVITQu4vr4xnSDxMaL",
            audio_effects=[scenario.effects.phone_quality()],
        ),
        scenario.JudgeAgent(criteria=[
            "Acknowledged the frustration before logistics",
            "Verified identity before any account action",
        ]),
    ],
    script=[scenario.agent(), scenario.user(), scenario.proceed(turns=5), scenario.judge()],
)
assert result.success
3

Run it

Run your usual test command (pytest or vitest). With LANGWATCH_API_KEY set, the run streams to the Simulations dashboard with full audio playback and per-segment transcripts, and writes a recordings/<scenario>/full.wav to listen back.
Voice scenarios are slower than text — TTS + transport + multi-turn means 30–120s per run, so give your test runner a generous timeout.

​
Dig into the full docs

Voice quick start

The complete five-minute walkthrough with a worked example per adapter.

Audio effects & interruptions

Background noise, codec degradation, custom WAVs, and barge-in recipes.

Capability matrix

Exactly which features each adapter supports.

Runnable examples

Demos per adapter and use case on GitHub.