vox

Name: vox
Availability: InStock
Author: Trevon

Provide local text-to-speech and speech-to-text over the MCP protocol.

1|1|Updated Feb 7, 2026

Search Tags:#speech to text #text to speech #on-device inference #voice mcp #audio dsp calibration #async listening #local realtime tools

Ready to run

npx skills add https://github.com/trvon/vox --skill vox

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: vox
Source: https://github.com/trvon/vox/tree/main
Command: npx skills add https://github.com/trvon/vox --skill vox

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Vox solves the problem of adding low-latency, local speech interfaces (text-to-speech and speech-to-text) to your workflow without cloud APIs.

Core Features & Use Cases

Local TTS + STT via MCP tools: Use say/enqueue_say to speak and listen/converse to transcribe what you say in real time.
Background listening + inbox retrieval: Use start_listening, then poll check_inbox to retrieve transcriptions asynchronously.
On-device DSP calibration for better accuracy: Use calibrate, plus reset_dsp, to tune microphone processing to your room.

Quick Start

Use the Vox skill to run the MCP tool listen so it records from your microphone and returns the transcribed text.

Frequently Asked Questions about vox

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

FAQPage Schema

How do I run speech to text and text to speech locally without cloud APIs?▼

You can run speech to text and text to speech locally by using on-device inference engines like Kokoro TTS and Moonshine Base STT over the MCP protocol, eliminating the need for cloud APIs and reducing latency.

Can I listen to my microphone in the background and retrieve transcriptions later?▼

Yes, you can start background listening with the start_listening tool and poll the check_inbox tool to asynchronously retrieve transcriptions from a queue, enabling non-blocking voice workflows.

How do I calibrate audio DSP for better speech to text accuracy in a noisy room?▼

You can calibrate audio DSP to improve microphone signal-to-noise ratio conditions by using the calibrate tool to tune processing for your specific room, with an option to reset_dsp if needed.

What is the best way to build an interactive speak-and-respond voice workflow on-device?▼

The best way to build an interactive speak-and-respond workflow is by invoking deterministic MCP tool handlers like say or enqueue_say for text to speech, and listen or converse for real time speech to text.

Does on-device voice transcription require any specific dependencies or cloud services?▼

On-device voice transcription requires no cloud services or external dependencies, operating entirely locally using structured JSON parameters and returning transcription or TTS queue status results.

Why use local realtime tools for text to speech instead of cloud-based voice APIs?▼

Local realtime tools for text to speech provide low-latency voice interfaces without cloud APIs, ensuring privacy and offline capability by processing audio entirely on-device through the MCP protocol.