videogame.io

Tools / Game Voice Maker

Game Voice Maker

Write a character and their lines; every line comes back spoken in that character's own steady voice, as MP3s the game plays.

Working - admin onlyComing soon for players
$0per line: no per-use fees
~1xreal time: a 4-second line takes about 4 seconds
1voice design per character, reused for every line
191lines voiced for one adventure game, 5 characters

How it works

The steps

  1. Give it the game's script. Qwen 3.8 picks the lines that should be spoken, names who says each one, and writes one voice description per character: age, accent, pitch, mood and pace.
  2. Ming TTS designs each voice once on a plain sentence, then clones that voice for every line, so a character sounds the same in line 1 and line 100.
  3. Numbers and short forms are rewritten to be read aloud first ("2 AM" becomes "two A M").
  4. Each line becomes an MP3 keyed to its on-screen text, and the game plays it while the line shows. Re-running records only new or changed lines; one line can be re-rolled on its own.
  5. Optional: a talking portrait (MuseTalk 1.5) lip-syncs the character's picture to the line, about 1-2 s of work per line.

Example 1 - HONEYWOOD GUARD

Elara, the quest giver (10 lines)

One voice description - "a bright young village woman in her early twenties, warm and quick, a little mischievous, light British lilt" - and every quest line read in it.

Example 2 - SCRIPT RUNNER

Dee's phone call (3 lines)

"A warm young woman in her twenties, American, relaxed and a little teasing, unhurried late-night phone voice." The game plays these when the phone rings.

Example 3 - Night Shift

A whole cast, with talking portraits

191 lines across 5 characters, each with its own voice. Press play: the portrait lip-syncs to the line.

Models and licences

What it runs on

StepModel or toolLicence
VoiceMing-3.0-tts-0.5B (inclusionAI)MIT (code and weights)
Pick lines, cast voicesQwen 3.8 Flash Nextopen weights (Qwen)
Read numbers aloudour English normaliserours
Talking portrait (optional)MuseTalk 1.5MIT (code and weights)

Voice cloning is only for a person's OWN recorded voice, with a consent box. Seven languages; tags such as [Laughter] and [Sigh] work on cloned voices.

Benchmarks and live pages