golden-voice

01 / THE OUTPUT

What it does

Record 30 seconds of yourself talking. golden-voice clones your voice locally using XTTS v2 and gives you one-command text-to-speech that sounds like you. No cloud, no API keys, no monthly bill.

# teach it your voice
golden-voice setup my-voice-sample.wav

# speak a line
golden-voice speak "The deployment looks clean. Ship it."

# pipe in a file
cat summary.md | golden-voice speak --stdin

02 / THE SCAR

Why it exists

We were paying ElevenLabs $100 a month for voice narration in our developer tools. Multiple sessions burned through the quota fast. The quality was good; the cost didn’t scale, and we wanted something we owned.

So we built it. One afternoon, zero budget. Now every developer on the team can keep a voice clone on their own laptop.

03 / THE MECHANISM

How it works

1. 01

   You record 15–30 seconds of yourself talking naturally.

2. 02

   XTTS v2 analyzes your voice—pitch, cadence, and timbre.

3. 03

   Any text you pipe in gets synthesized locally on your CPU.

4. 04

   Optional sox effects add reverb, chorus, or echo.

Generation takes ~15–20 seconds on Apple Silicon. For immediate feedback, --quick falls back to macOS say with warm voices.

04 / INSTALL

Three commands.
Then it’s yours.

git clone https://github.com/goldenfocus/golden-cloud.git
cd golden-cloud/blocks/golden-voice
bash install.sh

05 / SAMPLE LENGTH

Quality, honestly.

More clean speech gives the model more of you. Thirty seconds is the useful sweet spot.

06 / REQUIREMENTS

Before you run it

• 01macOS—Apple Silicon or Intel
• 02Python 3.11—the installer handles it through Homebrew
• 03About 2 GB of disk space for the one-time XTTS model download
• 04A clean 15–30 second recording of your voice