Shop / New · House Voice

House Voice

Give your local companion a real voice — running entirely on your own machine. A self-hosted neural TTS server (the open-source Kokoro model) plus a speak client with an OS-native fallback, so a line is never silently dropped. No cloud, no API keys, no per-character billing.

$24once · lifetime updatesStandalone source-code kit
A local neural voice, a cross-platform speak client, and an OS-native safety net

01In plain English

House Voice gives a locally-run companion a real voice on your own hardware — no cloud, no API keys, no per-character billing, and the audio never leaves your box. It's two small moving parts: a server (a ~200-line Python daemon wrapping the open-source Kokoro model, on 127.0.0.1:8768) and a client (speak.mjs — text in, audio out) you run as a CLI or import in one line.

The client's promise is that it never goes silent: if the neural server is unreachable, it automatically falls back to your operating system's built-in voice, so the line still speaks. The server's HTTP shape matches /v1/audio/speech, so many existing clients work unchanged. It ships no model weights (Kokoro fetches them from its official source on first run), no API keys, and zero npm dependencies on the client — Node built-ins only.

02What it looks like running

Start the server once, then speak a line — or pick a different voice, blend two, or save straight to a WAV. No server running? The same command still speaks, in your OS voice.

$ node speak/speak.mjs "Hello. This is my local voice." engine Kokoro · 127.0.0.1:8768 → played 1.8s of audio $ node speak/speak.mjs --voice am_adam --speed 1.1 "A touch faster." $ node speak/speak.mjs --out hello.wav "Save me to a file instead." # server offline? fallback OS voice (SAPI / say / espeak) → line still spoke, logged $ npm run verify → structural checks green

03What you need

This is a source-code package for builders, not a consumer app. You run it on your own machine — we ship the plumbing and the judgment; you install the model.

A coding agent

You don't have to be a coder — but you will need a coding agent to do the setup with you: Claude Code, Codex, or Cursor. It ships an agent-ready start-here you hand straight over, plus a full Kokoro install guide.

Node + (optional) Python

Node.js 18+ runs the speak client and self-test. Python 3.10+ runs the neural voice server — optional: skip it and the kit uses your OS's built-in voice. Runs on Windows, macOS & Linux. No accounts, no keys, no cloud.

Every Windows file has a Mac/Linux twin. The SAPI fallback voice (speak\fallback-sapi.ps1) and the server launcher (server\start-tts-server.bat) are [WINDOWS ONLY]; macOS/Linux use fallback-say.sh (say / espeak) and start-tts-server.command.

04Exactly what's in the kit

SinAndSarah-house-voice.zip
SinAndSarah-house-voice/
  server/  tts_server.py  requirements.txt  start-tts-server.bat [WINDOWS] start-tts-server.command
  speak/  speak.mjs  fallback-sapi.ps1 [WINDOWS]  fallback-say.sh
  config/  voice-config.json
  examples/  say-hello.bat  say-hello.sh
  docs/  INSTALL-KOKORO.md  LICENSES.md
  install.bat  install.sh  verify.mjs  .env.template
  VOICE_PLAYBOOK.md  BUYER-START-HERE.md  PRODUCT-MAP.md  README.md  LICENSE.txt

Ready when you are.

One-time purchase, lifetime updates. Checkout is Stripe — secure card payment, then an instant download plus your link by email.

Buy House Voice — $24