voice-e2e

Name: voice-e2e
Availability: InStock
Author: lovon-spec

Automate end-to-end audio testing of live voice pipelines via browser and virtual audio hardware.

2|Updated Mar 2, 2026

Search Tags:#audio #e2e-testing #firefox #xvfb #pulse-audio #virtual-mic

Ready to run

npx skills add https://github.com/lovon-spec/intendant --skill voice-e2e

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: voice-e2e
Source: https://github.com/lovon-spec/intendant/tree/main/tests/skills/voice-e2e
Command: npx skills add https://github.com/lovon-spec/intendant --skill voice-e2e

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This skill enables end-to-end testing of the live audio pipeline for AI agents by simulating real audio through the full stack: espeak-ng TTS, a PulseAudio virtual microphone, Firefox getUserMedia, the AudioWorklet/WASM layer, and the live model, with human monitoring via VNC.

Core Features & Use Cases

End-to-end audio path testing across TTS, virtual microphone, browser capture, and live model interaction.
Validation via a debug endpoint and VNC-based monitoring to ensure the pipeline stays connected and responsive.
Useful for CI workflows to verify audio paths and model responses before release, especially for Gemini Live and OpenAI Realtime integrations.

Quick Start

Run the voice-e2e workflow to verify the full audio capture and live model path in your environment.

Frequently Asked Questions about voice-e2e

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I automate end-to-end audio testing for a live voice pipeline in a headless browser environment?▼

Automate end-to-end audio testing by routing synthesized speech through a virtual microphone into Firefox getUserMedia, verifying TTS, WASM processing, and live model interactions via a debug endpoint. This pipeline uses Xvfb and PulseAudio to simulate real audio capture without a physical microphone.

What prerequisites are needed to test browser-based audio capture with a virtual microphone?▼

Testing browser-based audio capture requires espeak-ng for TTS, PulseAudio for virtual audio routing, Xvfb for headless display, and Firefox for getUserMedia. These components establish the virtual hardware layer needed to simulate live audio input.

Can I use Xvfb and PulseAudio to simulate real audio input for CI workflows?▼

Yes, Xvfb and PulseAudio simulate real audio input for CI workflows by creating a virtual display and routing synthesized audio to a virtual microphone. This validates audio paths and live model responses before release without manual interaction.

Why does getUserMedia fail to capture audio during automated Firefox testing?▼

getUserMedia fails to capture audio when PulseAudio virtual hardware is misconfigured or Xvfb lacks display access. Properly mapping the virtual microphone source to the browser ensures the AudioWorklet receives the simulated TTS stream.

Does end-to-end voice testing work with OpenAI Realtime and Gemini Live integrations?▼

End-to-end voice testing works with Gemini Live and OpenAI Realtime integrations by exercising the full audio capture and live model interaction path. It verifies audio transmission and model responsiveness across the complete pipeline.