genpark-voice-vad

io.github.Alpha-ParkPublished by GitHub account Alpha-Parkv1.0.1Updated Oct 24, 2026

Energy and transcript heuristics for voice turn endpoint detection.

VerifiedSTDIODesktop onlyMedia & DesignAI & ML

Overview

AI-generated overview

A local MCP server that uses energy and transcript heuristics to judge whether a speaker has finished their turn in voice conversations.

What you can do
Analyze turn status from precomputed dB energies and a transcript to decide if a speaker has stopped talking or is still mid-turn. Calibrate acoustic thresholds for the energy heuristics. Predict semantic closure from the transcript. Run a benchmark of turn detection on isolated instances.
When to use it
Useful when building or testing voice turn-taking logic and you already have precomputed energy values and a transcript. Not suitable for decoding audio, transcribing speech, or running a trained VAD model.
Requirements
Local Python 3.9 or newer; the library and stdio server have no runtime dependencies. Install from the GitHub release wheel or the .mcpb desktop bundle, which needs a Python 3.9+ interpreter on PATH. The PyPI package is not yet published.
Before you install
Confidence values are heuristic scores, not calibrated probabilities. It does not process raw audio or transcribe speech, so inputs must be precomputed energies and a transcript. Each MCP process keeps its own state.

Installation

In SourceWeft

  1. Open genpark-voice-vad in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

genpark-voice-vad

Energy and transcript heuristics for voice turn endpoint detection.

This is a heuristic endpoint detector over precomputed dB energies and a transcript. It does not decode audio, transcribe speech, or run a trained VAD model. Confidence values are heuristic scores, not calibrated probabilities.

Install from the GitHub release

Python 3.9 or newer. The library and stdio MCP server have no runtime dependencies.

sh
python -m pip install https://github.com/Alpha-Park/genpark-voice-turn-taking-endpoint-detector-skill/releases/download/v1.0.1/genpark_voice_vad-1.0.1-py3-none-any.whl

PyPI publication is pending account setup. The intended PyPI project is genpark-voice-vad; do not assume pip install genpark-voice-vad is available until the project is published.

Python usage

python
from genpark_voice_vad import VoiceTurnTakingEndpointDetectorclient = VoiceTurnTakingEndpointDetector()print(client.run_benchmark_turn_detection())

MCP stdio configuration

After installing the wheel, configure your MCP client with the installed command:

json
{  "mcpServers": {    "genpark-voice-vad": {      "command": "genpark-voice-vad",      "args": []    }  }}

If the command is not on PATH, use its absolute path or python -m genpark_voice_vad with the same interpreter where you installed the wheel. The GitHub release also contains a .mcpb bundle for clients supporting desktop extensions. That bundle requires a Python 3.9+ interpreter on PATH; it bundles the server source.

Available tools: analyze_turn_status, calibrate_acoustic_thresholds, predict_semantic_closure, run_benchmark_turn_detection. tools/list returns required arguments and JSON schemas. Each MCP process holds its own state. Benchmark tools use isolated instances.

Development

sh
python -m unittest discover -s testspython -m pip install mcppython tests/check_mcp.pypython -m pip install build twinepython -m buildpython -m twine check dist/*

python mcp_server.py --test runs the deterministic example; it is not a protocol conformance test. The MCP client check exercises initialize, tools/list, tools/call and ping over stdio.

Distribution

GitHub source and release artifacts are the primary distribution until PyPI is configured. Registry submissions are tracked separately; a manifest is not proof of registry acceptance. See PUBLISHING.md for the repeatable PyPI workflow.

MIT license. Maintained by GenPark.

Source: README.md at commit 69cdc3d

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

1
  1. v1.0.1LatestSep 27, 2026

Score

40% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts

    Declare a license in the repository or the registry entry.