Voice cloning lets you record a short sample of your voice once, then narrate unlimited scripts in that exact voice — no re-recording. The catch with most tools is that you upload your voice to someone else’s server. This guide shows how to clone your voice for videos entirely on-device, so the reference audio never leaves your machine, using the free, open-source ViralMint desktop app.

By the end you’ll have a saved voice you can pick anywhere you’d normally choose a text-to-speech voice, and a finished MP4 narrated in it.

What “on-device voice cloning” actually means

Most AI voice cloning services send your reference clip to the cloud, train a voice on their servers, and stream synthesized audio back. That works, but your voiceprint now lives on a third party’s infrastructure.

ViralMint clones your voice locally instead. It runs the open-source VoxCPM engine (VoxCPM 0.5B via the Apache-2.0 VoxCPM.cpp port) directly on your computer. The reference audio and every second of synthesis happen on-device — nothing is uploaded. That’s the core privacy differentiator: your voice is data you keep, not data you hand over.

Two honest constraints before you start:

  • Platform: on-device cloning runs on macOS with Apple Silicon (M-series) and Linux only — not Windows or Intel Macs yet.
  • First run: the plugin downloads a small native binary plus a ~545MB model the first time you use it. After that it’s fully offline.
  • Ethics: clone your own voice, or one you have explicit rights to. Don’t clone anyone without consent.

How to clone your voice — step by step

The whole flow takes about 15 minutes, most of which is the one-time model download.

Step 1: Install ViralMint and the voice-cloning plugin

Download ViralMint — it’s free and open-source (AGPL-3.0), a real desktop app rather than a browser tab. Install it on a Mac with Apple Silicon or a Linux machine.

Open the app and go to Tools → Voice Cloning (route /tools/voice-cloning). The first time you open it, ViralMint fetches the on-demand plugin: a small native binary and the ~545MB VoxCPM model. Let it finish once and you won’t wait again.

Step 2: Record or upload a 10–30 second reference clip

You need one clean sample of the voice you’re cloning. Either:

  • Record directly in the voice-cloning page, or
  • Upload an existing clip (a clip of yourself talking to camera works fine).

Aim for 10–30 seconds, a single speaker, in a quiet room. No background music, no overlapping voices, minimal room echo. A short, clean sample clones better than a long, noisy one — quality of the reference is the single biggest factor in how natural the clone sounds.

Step 3: Let it auto-transcribe and name the voice

Cloning works best when the model knows exactly what words are in the reference clip. ViralMint handles that for you: it transcribes the sample locally with Whisper to produce the matching transcript, so you don’t type anything by hand. (You can edit the transcript if Whisper mishears a word.)

Give the voice a name — “My narration voice”, your channel name, whatever you’ll recognize — and save it. It now lives in your voice library.

Step 4: Narrate any script in your cloned voice

Your saved clone is selectable anywhere ViralMint offers a voice: the Voice-over tool, Smart Video, and other creation flows. Pick it exactly like you’d pick any text-to-speech voice.

Paste or write a script and synthesize. A note on speed: synthesis runs on your CPU at roughly 5× realtime — a 30-second voiceover takes about 2.5 minutes to render — so ViralMint runs it as a background job rather than freezing the screen. Kick it off and keep working.

Step 5: Export the finished MP4

If you’re building a full short in Smart Video, ViralMint times word-by-word captions to your cloned narration automatically and can drop in matched b-roll. Export when you’re happy — you get a plain MP4 with no watermark, which you download and post yourself. ViralMint never auto-uploads anything.

What it costs

Cloning the voice and running the local engine is free — there’s no per-clone provider cost because the model runs on your hardware. Actually synthesizing a voiceover is pay-as-you-go: 5¢ per 1,000 characters, capped at 20¢ per generation. A free ViralMint account also comes with a small daily allowance to use before you top up any prepaid balance. No subscription, ever.

That’s cheaper than most cloud voice-cloning subscriptions, and the money goes to compute you’re barely using — the synthesis itself is local.

When to clone vs. use a stock AI voice

Cloning is worth it when your voice is part of your brand — you want consistency across a series without re-recording every script, or you’re localizing your own content and want it to still sound like you. For contrast:

  • Free stock voices — ViralMint also ships free Edge TTS (400+ voices, 100+ languages, cloud-Microsoft, not cloning). Great when any clean voice will do. See the best free AI voice generator.
  • Premium AI voices — Gemini 3.1 Flash TTS at $0.12/1,000 characters for extra-expressive stock narration.
  • Your cloned voice — when it specifically has to sound like you, kept private on your own machine.

Frequently asked

Is AI voice cloning free? Cloning and running the on-device engine is free; synthesizing is 5¢/1,000 characters capped at 20¢ per generation, with a small free daily allowance. No subscription, no watermark.

Does my audio get uploaded? No. The reference clip and all synthesis run 100% locally via the open-source VoxCPM engine — your voice never leaves your machine.

Which OS do I need? macOS with Apple Silicon or Linux. Windows and Intel Macs aren’t supported for cloning yet.

How long a sample? A clean 10–30 second clip of one speaker in a quiet room.

Can I clone any voice? Only your own voice or one you have explicit rights to use — never someone else’s without consent.