Voice cloning explained · 6:41

How voice cloning works and how to make a presentation in your own voice

See how a short reference sample guides new speech, then follow a practical Mac workflow that turns a timed script and three slides into a narrated presentation.

Published September 18, 2026 · Presented by Vincent Chu · Watch on YouTube

What the video covers

From voice reference to finished narration.

The first half explains the difference between a model’s original training and the short voice sample you provide later. It covers speaker conditioning, audio representations, zero-shot cloning, and why the result still needs a careful listen.

The second half uses the Mac command line included with SpeechTrack: Live Voice Clone. It selects a saved voice, reads timed narration from CSV, renders an audio track, and combines that track with three slide images using FFmpeg.

Video chapters

Jump to the part you need.

0:15What makes a voice recognizablePitch is only one part of a speaker’s sound. 0:31Training and voice conditioningHow an existing model uses a new reference sample. 1:20Pocket TTS and zero-shot cloningA practical view of latents, generation, and decoding. 2:14Record a useful voice sampleRoom, microphone, delivery, consent, and review. 2:50Build the presentation narrationSaved voices, the Mac CLI, CSV timing, and audio export. 5:09Combine slides and speechUse FFmpeg to create the final narrated presentation.

Made with SpeechTrack: Live Voice Clone

Create timed speech locally on your Mac.

Record or import a permitted voice sample, arrange script segments on a timeline, and export one narration track. Voice samples, scripts, and generation stay on your device.

Get SpeechTrack: Live Voice Clone Learn about the app