Field notes from the studio

The Vocast blog.

How we build voice cloning and narration that you can measure. Methodology, metrics, local-first engineering, and product notes.

MethodologyFeatured · 6 min read

Natural enough to publish: an AI voice you can actually use

Most AI voices sound great in a ten-second demo and fall apart over twenty minutes. Here is how we measure naturalness, and what it takes to ship a cloned voice you would put your name on.

Minhyeok Kang & Vocast Team· Jul 19, 2026
Engineering
Measuring prosody, not vibes
Why Vocast scores pitch, pauses and sentence endings separately, and gates every render on all of them.
Vocast Team· Jul 12, 20266 min read
Voice
Clone your voice in ninety seconds
Ten guided lines, one reusable profile you can version and roll back. Your voice as an asset you own.
Minhyeok Kang· Jul 4, 20265 min read
Engineering
Long-form without the re-record
20,000-character scripts, per-paragraph regeneration, and performance transfer for the passages that need a specific delivery.
Vocast Team· Jun 26, 20265 min read
Privacy
Why Vocast is one-time and fully local
No subscription, no server, no account. Your voice and your words stay on your machine, by design.
Minhyeok Kang· Jun 18, 20263 min read
Voice
How to fix one bad line without re-recording the whole take
One stumbled line should not cost you a whole session. Why a single line is hard to replace, how far editing gets you, and when regenerating beats re-recording.
Minhyeok Kang· Jul 25, 20265 min read
Engineering
Why your denoiser swallows word endings, and how to measure it
Aggressive noise reduction files the tails off your words. Here is why classic denoisers do it, how to measure the loss instead of guessing, and how to clean without the damage.
Vocast Team· Jul 24, 20265 min read