Telephony Voice SimulatorTelephonyVoiceSimulator

Telephony Voice Simulator tests the boundary where voice agents meet the real phone network. It brings answering-machine detection, inbound IVRs, PBX routing, queues, screeners, human pickup, recordings, and grading into one repeatable workspace.

Voice systems fail in ways that text-based tests do not capture: unusual voicemail greetings, early beeps, silence, DTMF gates, carrier errors, queue overflow, and timing-sensitive transfers. The simulator makes those conditions reproducible before they affect live callers.

The platform includes more than forty scenarios across three domains:

• Answering-machine and human-pickup behavior
• IVR routing, invalid input, retry, silence, busy, and abandonment
• PBX schedules, extensions, queues, overflow, fallback, and voicemail

A Python control plane owns scenarios, provider profiles, runs, recordings, grading, and analytics. A Next.js console provides the operator experience, while adapters connect local simulation to Twilio and LiveKit.

The default mode runs locally with SQLite. Controlled ingress exposes only signed carrier webhooks and reviewed media routes for live Twilio testing; the private control API stays on loopback or a protected network.

Multi-step UAT flows can provision a scenario, initiate a call, capture values, retry transient failures, grade the outcome, and collect artifacts. That makes voice-agent regression testing portable across runtimes and providers.

Recordings are opt-in, provider data is isolated, and public builds exclude restricted audio corpora. The repository documents consent, retention, deletion, and provider-hosted recording boundaries rather than treating call capture as a default.