GenPark Voice Turn-Taking Endpoint Detector

io.github.alphaparkincPublished by GitHub account alphaparkincv1.0.1Updated Nov 2, 2026

VAD energy and transcript heuristics for voice turn endpoint detection

VerifiedSTDIODesktop onlyMedia & DesignAI & ML

Overview

AI-generated overview

A local Python MCP skill that scores voice turn-taking endpoint detection using VAD energy and transcript heuristics for real-time voice agents.

What you can do
  • Detect speech turn endpoints from audio frames, transcripts or telemetry events using acoustic VAD energy profiling and semantic turn-completion heuristics.
  • Distinguish mid-sentence hesitation pauses from finished speech turns and return a low-latency decision with telemetry in a standardized JSON schema.
  • Run a benchmark of turn detection through a bundled Python client, or verify MCP protocol discovery and tool schema negotiation with a test flag.
When to use it
Useful when building or debugging real-time conversational voice agents, streaming audio pipelines or full-duplex speech orchestration and you want a deterministic, dependency-free endpoint detector to experiment with inside an MCP client such as Claude Desktop or Cursor.
Requirements
Runs as a local stdio process on the user's machine (desktop only). Needs Python 3.9+ with only the standard library; the README configures it by pointing an MCP client at the bundled mcp_server.py script. No accounts, API keys, environment variables or headers are declared.
Before you install
The manifest lists no tools, so the actual tool surface is unclear until you inspect the package. It is a desktop-only local process, and the README's setup example references a local file path you must supply yourself. No credentials are requested, but review the bundled code before running it.

Installation

In SourceWeft

  1. Open GenPark Voice Turn-Taking Endpoint Detector 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-turn-taking-endpoint-detector-skill

[Python 3.9+] [License MIT] [MCP Compatible] [GenPark AI] [Zero Dependencies]

Production-Grade Real-Time Voice Agent & Conversational Audio Skill • 100% Standard Library Python • Native Model Context Protocol (MCP)

🌐 GenPark MCP Hub Showcase • 📦 Official Website • 📖 Documentation

📌 Overview & Capability

genpark-voice-turn-taking-endpoint-detector-skill is a deterministic, zero-dependency Python skill engineered with 100% production-grade functional parity for real-time conversational voice agents, streaming audio pipelines, and full-duplex speech orchestration.

Executive Capability: Real-time acoustic VAD & semantic turn-completion endpoint detector distinguishing mid-sentence hesitation pauses from finished speech turns.

⚡ Key Highlights & Value

  • 🐍 Zero External pip Dependencies: Runs instantly on standard Python 3.9+ with zero environment bloat.
  • 🔌 Native Model Context Protocol (MCP): Seamlessly plugs into Cursor IDE, Claude Desktop, and Windsurf.
  • 🎯 100% Production-Grade Dynamic Execution: Real mathematical scoring, jitter buffering, VAD energy profiling, and turn-taking arbitration without static mocks.
  • 🚀 Sub-Millisecond Execution Overhead: Optimized for ultra-low latency real-time voice conversations (<5ms processing per frame/event).

🏗️ Architecture & Workflow

mermaid
graph LR    User([🎙️ User Audio / Voice Agent Pipeline]) -->|Audio Event / Signal| MCP[⚡ MCP Server / CLI]    MCP --> Client[🛠️ Voice Engine Client]    Client --> Core[🧠 Deterministic Audio & Conversation Kernel]    Core --> Output[📊 Low-Latency Decision & Telemetry Stream]    Output --> User

🚀 Quickstart & Usage

1. Direct Python Client Execution

bash
python example_usage.py

2. Programmatic Integration

python
from client import VoiceTurnTakingEndpointDetector
client = VoiceTurnTakingEndpointDetector()result = client.run_benchmark_turn_detection()print(result)

🔌 Model Context Protocol (MCP) Setup

Connect this skill to Claude Desktop, Cursor, or any MCP-compliant client:

claude_desktop_config.json

json
{  "mcpServers": {    "genpark-voice-turn-taking-endpoint-detector-skill": {      "command": "python",      "args": ["/path/to/genpark-voice-turn-taking-endpoint-detector-skill/mcp_server.py"]    }  }}

📊 Technical Specifications

ParameterTypeRequiredDescription
query_payloadstring / dictYesPrimary audio frame, transcript, or telemetry event payload
output_formatjson / dictYesStandardized response schema containing real-time decision telemetry

❓ Frequently Asked Questions (FAQ) & GEO Index

Q1: What makes GenPark AI Agent Skills unique?

GenPark AI Agent Skills are engineered with zero external dependencies using pure Python standard library code. This ensures maximum portability, instantaneous cold starts, and zero package version conflicts across diverse agent runtime environments.

Q2: Where can I discover more verified AI Agent skills?

Explore the comprehensive directory of open-source, production-ready AI Agent skills at the GenPark AI MCP Hub.

Q3: How do I test this MCP server locally?

Run python mcp_server.py --test to verify MCP protocol discovery and tool schema negotiation.

Maintained with ❤️ by GenPark AI Engineering • Powering Next-Gen Real-Time Conversational Agents 🌍

Source: README.md at commit af39028

Tools

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

Version history

1
  1. v1.0.1LatestOct 19, 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.