About Services Tools Blog Contact us

Voice Cloning from Reference Clips

Builds a reusable speaking voice from short reference clips and reads later scripts in that voice, entirely on the studio machine.

  • Voice Cloning
  • Media production
  • Tool
  • Productised
  • Local/private
  • 2026

The problem

A consistent narrator meant booking the same speaker every time a script changed. A flat stock voice is faster, but it does not sound like the person the audience already knows.

What it does

You give it short clean clips and a voice name. It checks the clips for quality, transcribes them and saves the voice. Later you give it a script and it speaks the lines in that voice and saves an audio file. The file is kept until the script changes. The reference recordings stay on the studio machine, and the sample on this page is a cloned reading of an invented script in the house voice.

Outcomes

  • A named voice can read a revised script without a new recording session.
  • An audio file sits beside the script, ready to drop into an edit.
  • An unchanged line is not generated again because the audio is saved.
  • A folder of texts can run as a batch where the extra tools are installed.

Simplified mockup

Interactive preview with sample data. Nothing is sent to a server.

Reference clips

3 files

Folder sketch for voices/studio-narrator/. These cards are not the published recordings. The cloned read is on Speak.

Mean quality
0.81
Longest clip
8.2 s
Weakest
0.61
still usable
ref-a.wav 0.93

Clean, low noise. 8.2 s.

Sibling: “Start on the wide shot, then come in.”

ref-b.wav 0.88

Clean, a little room. 6.4 s.

Sibling: “Hold for a beat, then read the line.”

ref-c.wav 0.61

Background rustle. Kept, not dropped. 4.1 s.

Sibling: “This take has a little room in it.”

Register a voice

Profile

A plain name for the folder. Reference clips stay on this machine. Speak plays the published clone.

Build
Ready to register
Not registered Three clips are scored. Register writes the profile and the status line.

Read a line

Studio Narrator
Script

This is the sentence the published file reads. An unchanged script keeps the existing recording.

Cache
cloned-read.mp3 matches this sentence
Ready to reuse the cached clip
cloned-read.mp3 Studio Narrator House voice

A cloned read of new words. The reference recordings are not in this file.

How it works

  1. Place reference clips in one folder per voice, or cut a short speech section from a longer recording.
  2. Check the sample quality and transcribe the clips so the voice profile has an accurate written copy.
  3. Save the voice under a plain name.
  4. Turn the supplied script into speech with a voice cloning model and save an audio file.
  5. Reuse the existing audio file when the script text has not changed.

Pipeline

flowchart LR
  Clips[Reference clips] --> Score[Quality score]
  Score --> Profile[Voice profile]
  Profile --> Script[Script text]
  Script --> Synth[Higgs Audio]
  Synth --> Wav[Narration WAV]

Sample output

Cloned read of new words in the house voice