On Patreon

TTS Lab

A standalone, local playground for Supertonic voices. Type a line, pick a voice, hear it, fast: then shape it, blend voices into new ones, and save the line as a WAV.

Runs locally, WindowsVoices 10, plus your ownExport WAV

Get it on Patreon

$ tts-lab
  voice   Male 3  ·  mood Gruff
  speed   0.9x  ·  pitch −2  ·  formant −1
  fx      Radio
› "Ye'll not cross this bridge, traveller."
  ▸ first sound in 212 ms
  ▸ saved line_014.wav

A LINE IN THE MAKING

What it does

Shape a voice until it fits

Speak

Hear it as it’s made

Ten stock Supertonic voices, five female and five male
Audio streams as it’s made, with the time to first sound shown
Recent lines kept in your browser to replay or download
Accents by spelling: try Cockney light or heavy

Shape

Every dial you need

Speed from 0.7x to 2x
Mood: normal, upbeat, gruff, meek or intense
Pitch and tone, ±6; formant, ±4
Character effects: telephone, radio, lo-fi
Quality steps, and the pause between phrases

Blend

Make new voices

Mix stock voices by weight, add effects, and save the mix as a voice of its own. New voices show up straight away.

Build

A voice from a recording

Upload a short recording (WAV, MP3, OGG, FLAC or M4A) or point at a link, and build a voice from it. Needs an NVIDIA graphics card.

Get started

Install

You need

Requirements

Windows and Python
About 400 MB for the voice model, downloaded on the first run
An NVIDIA GPU only for building voices from recordings
About 10 to 15 minutes to set up

Steps

Up and running

python -m venv .venv
.venv\Scripts\python.exe -m pip install -r requirements.txt
Run run.bat; it opens http://localhost:8020.
stop.bat stops it.

The voices are Supertone’s Supertonic 3 model, under its OpenRAIL-M licence, downloaded on first run.

Get TTS Lab

The release is on Patreon.

Patreon