# genpark-voice-vad

Decides whether a speaker has finished their turn, using precomputed dB energies plus a transcript. Useful in voice bots and dialogue pipelines.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed · ★ 7
- Author: Alpha-Park
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/Alpha-Park/genpark-voice-turn-taking-endpoint-detector-skill
- Updated: 2026-09-28
- HTML: https://affpapa.org/en/nevibecode/mcp/genpark-voice-vad

## What it does

The server reads dB energy values and a transcript that you already have. It never decodes audio, never transcribes speech, and runs no trained VAD model. Confidence numbers are heuristic scores rather than calibrated probabilities, so feeding them into thresholds tuned for a real model will mislead you.

Four tools ship with it: `analyze_turn_status`, `calibrate_acoustic_thresholds`, `predict_semantic_closure` and `run_benchmark_turn_detection`. Calling `tools/list` returns the required arguments and their JSON schemas. State lives per MCP process, and benchmark tools use isolated instances.

Python 3.9 or newer is required. Neither the library nor the stdio server pulls in runtime dependencies. Install comes from the wheel attached to the GitHub release, since PyPI publication is still pending — `pip install genpark-voice-vad` is not available. When the `genpark-voice-vad` command is missing from `PATH`, point to its absolute path or run `python -m genpark_voice_vad` with the same interpreter you installed into.

## Requirements
- Needs: Python and uv (commands run via uvx)

## Install

### Claude Code (native platform)

**Install**
- The MCP registry has no launch command for it: see the repository README

## When to use it
- tell from a transcript whether someone finished speaking or is just pausing
- tune energy thresholds against a specific recording
- score semantic closure before the bot replies
- benchmark turn detection on your own data
- add endpoint detection to an MCP client without audio libraries

## Example prompts
- Check this transcript and energy series to see if the speaker finished their turn
- Calibrate the acoustic thresholds for this recording
- Run the turn detection benchmark on my test set

## Similar
- [MCP Toolbox for Databases](https://affpapa.org/en/nevibecode/mcp/googleapis-mcp-toolbox-for-databases.md) — MCP server · ★ 16717
- [labelhead-artist-momentum](https://affpapa.org/en/nevibecode/mcp/labelhead-artist-momentum.md) — MCP server · ★ 107423
- [Agent Skills Search Server](https://affpapa.org/en/nevibecode/mcp/agent-skills-search-server.md) — MCP server · ★ 26559
- [Skyvern](https://affpapa.org/en/nevibecode/mcp/skyvern.md) — MCP server · ★ 23263
- [Nhost](https://affpapa.org/en/nevibecode/mcp/nhost.md) — MCP server · ★ 9366
- [PraisonAI](https://affpapa.org/en/nevibecode/mcp/praisonai.md) — MCP server · ★ 9289

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md