# How to make a voiceover in your own voice

_By Apoorv Khanna, November 18, 2026_

**To make a voiceover in your own voice, either record it yourself and have an agent clean up the takes, or clone your voice once from about a minute of clean speech, with your consent, and generate new lines from text.** Recording is cheapest and sounds most like you. Cloning saves time when the script keeps changing.

## What people usually try, and where it falls short

- **One take on a phone.** Fast. Room echo, a fan in the background and a few "ums" make it sound rushed.
- **A stock AI voice.** Clear and quick, but it is not you. Fine for a demo, odd for a personal channel or a founder video.
- **Voice cloning apps.** ElevenLabs, Descript and others clone a voice from a sample, usually on a monthly plan.
- **Hiring a voice actor.** Great for ads. Not your voice either.

## Route 1: record it, and let the agent do the cleanup

1. **Write for speaking.** Ask your agent to shorten sentences and spell out numbers: "two hundred dollars", not "$200".
2. **Record in short sections.** One paragraph per take, in a small soft room, phone about a hand away and a little to the side.
3. **Check the words.** [deepgram/transcribe](/tools/deepgram/transcribe) turns each take into text for about 1¢ a minute, so the agent can flag missing or misread lines against the script.
4. **Remove background noise.** ElevenLabs' Voice Isolator, reached through `monid/run-metered`, strips fans, traffic and room hum. It is listed at 12¢ a call, billed on usage.
5. **Trim and join.** The agent cuts long pauses and false starts and joins the takes with ffmpeg on your machine.

This keeps your real voice. Every word was said by you.

## Route 2: clone your voice once

Cloning means recording a clean sample once, then typing new lines that come out in your voice.

**Record a good sample:**

- About a minute of natural speech. More than three minutes rarely helps.
- A small, soft room. A bedroom or a closet of clothes beats a kitchen. Echo gets cloned too.
- Fans, air conditioning and notifications off.
- Your on-camera voice, not a reading voice.
- No music, no other voices.

**Consent comes first.** Clone your own voice, or the voice of someone who agreed. Good services make the speaker read a consent line in their own voice. Keep the sample private. Label the voice as AI where a platform asks, and never use it to pass as someone in a call or a message.

Vaaya does not offer voice cloning today. [ThreeCut](https://threecut.com), a sister product, does. You record six short lines, starting with a consent line, and it checks each take for level and room noise as you go. A photo is optional: with one, your avatar appears on camera; without one, your voice narrates over the pictures. Type `@yourname` in a video's topic and the writer gives your voice the lines. Delete the avatar and its voice goes with it. Its [avatar guide](https://threecut.com/avatars/guide) covers the recording setup.

ThreeCut makes videos, so the voiceover arrives inside an MP4. If you only need the audio, one ffmpeg line pulls it out:

```text
ffmpeg -i video.mp4 -vn -c:a copy voiceover.m4a
```

## The prompt to copy for route 1

Run this in a coding agent with Vaaya connected, such as Claude Code, Cursor or Codex, in a folder with your takes and the script:

```text
I recorded my voiceover in takes: take-01.m4a, take-02.m4a, take-03.m4a.
The script is script.md.

Use Vaaya. For each take:
1. Transcribe it with deepgram/transcribe (set duration_minutes to the take's
   length) and compare it with the script. List any missing or wrong words.
2. Remove background noise with the ElevenLabs Voice Isolator through
   monid/run-metered. Ask consult for the exact input and price first.
Then cut silences longer than 0.7 seconds and join the cleaned takes in order
with ffmpeg into voiceover.mp3. Do not change my voice or pitch.
Tell me the total price before you start. Spend at most 75 cents.
```

Your agent will need the takes at a public URL for the transcription call. It can stage them with Vaaya's file upload, which consult will point to.

## What it costs

Prices as listed in the Vaaya catalog on 18 November 2026:

| Step | Tool | Price |
| --- | --- | --- |
| Check the words | `deepgram/transcribe` | about 1¢ a minute, 2¢ minimum |
| Remove noise | ElevenLabs Voice Isolator via `monid/run-metered` | listed at 12¢ a call |
| Stock-voice draft | `fal/generate`, ElevenLabs Turbo v2.5 | 5¢ per started 1,000 characters |
| Trim and join | ffmpeg on your machine | free |

A two-minute voiceover in three takes is about 6¢ for transcripts and about 36¢ for noise removal, so under 50¢. On ThreeCut, new accounts get 2,500 free credits, worth $25. A Faceless video starts at 1,400 credits and packs start at $20 for 2,000 credits, as listed on 18 November 2026.

## Common mistakes

- **Recording in a hard room.** Echo is the hardest problem to fix later, and a clone copies it.
- **Reading instead of talking.** Your reading voice sounds stiff. Speak to one person.
- **One long take.** Short sections are easier to redo.
- **Cloning someone without their OK.** A mutual friend saying "they won't mind" is not consent.
- **Skipping the listen.** A transcript confirms the words, not the tone. Play it through once before you publish.

To narrate with a stock voice instead, read [how to create a voiceover for a script](/blog/how-to-create-a-voiceover-with-vaaya). To make a two-host episode from a document, see [how to turn a PDF into a two-voice podcast](/blog/how-to-turn-a-pdf-into-a-podcast).

## Questions

**How long a voice sample do I need to clone my voice?**

About a minute of clean speech is enough for most current voice cloning tools. ThreeCut asks for about a minute and says more than three minutes does not make the voice better. A quiet, soft room matters more than length.

**Is it legal to clone a voice?**

Cloning your own voice is fine. Cloning someone else's needs their clear agreement, and many services make that person read a consent line themselves. Using a cloned voice to deceive or impersonate someone can break the law and platform rules.

**Does Vaaya have a voice cloning API?**

Not today. Vaaya has stock-voice text to speech, transcription and noise removal. For a voiceover in your cloned voice, ThreeCut, a sister product, builds a voice from about a minute of your speech.

**Should I tell viewers the voice is AI?**

If the voice is generated, yes, where the platform asks. YouTube, TikTok and Instagram all have AI disclosure settings for realistic synthetic content.
