REVIEWS / AI VOICE / OWNER INSIGHTS
🦉 WE READ 52 OWNER COMMENTS
Descript Overdub: what owners actually say
Owners find the voice cloning impressive and genuinely useful for content creation, but hit friction with training requirements, limited inflection control, and occasional UI bugs.
What owners complain about
- Not a perfect voice match SOME
Even with high-quality microphone training in a soundproofed closet, users report the cloned voice is impressive but 'not a perfect match for my actual voice.' One user noted their training audio sounded 'a little tinny,' likely affecting output quality.
- No inflection control SOME
Users want the ability to go back and adjust individual inflections after generation. One commenter explicitly said 'all this would need is a way to go back through and change a couple inflections,' and another noted you can record in different styles but hadn't had time to experiment with whether it helps.
- Rising training requirements FEW
The app now requires at least 10 minutes of audio to process, up from what was previously less. One user called trialing 'a bit more of a pain' after the change. Full training involves reading 30 minutes of Wizard of Oz, with an option for an additional supplemental hour.
- UI bugs and missing features SOME
Users report the waveform not showing after adding text, and not seeing the 'speakers and voices' panel next to their text. Others wish for mobile support and Arabic language reproduction, neither of which appear available.
- TOS grants broad usage rights FEW
The terms make it clear Descript will use everything you do to improve their product. One commenter noted it 'might make you think twice' about what you upload, though they acknowledged it isn't unreasonable for a free tier.
What owners love
- Genuinely impressive voice quality
Multiple users call the cloned and stock voices 'impressive,' 'extremely impressive,' and note the output is 44.1kHz and 'super crisp.' Several say it already beats bad human audiobook narration and sounds more natural than competing AI voice tools that produce 'robotic' results.
- Solves real content creation problems
A podcast host of 18 months discovered they could generate episodes read by their own cloned voice without hiring someone. A 3D animation hobbyist said voices were 'the biggest obstacle to overcome' and Overdub's demos were convincing enough to solve it. A virtual radio station owner uses it in production.
- Stock voices expand utility
Descript launched built-in stock voices created via the same cloning pipeline, giving users options beyond training their own voice. Commenters found these demo voices compelling for voice-overs and e-learning.
- Accessible entry point
Users appreciate being able to trial and experiment with the technology. One user enjoyed recording silly character voices and fake accents just to see what the merged result sounded like, calling it 'hilarious.'
Surprising patterns
- Training requires reading 30 minutes of Wizard of Oz specifically — users follow along with the script and record in improvised booths (closets with foam mattress pads and blankets).
- Recording training audio with multiple silly character voices and fake accents produces a single merged voice that sounds unintentionally hilarious — some users experiment with this deliberately.
- An 18-month paying customer didn't know Overdub existed until watching a tutorial video, suggesting the feature is under-discovered even among existing Descript users.
- Before Lyrebird was acquired by Descript, it was a free API endpoint that developers integrated into projects — those integrations silently broke after the acquisition, leaving developers looking for alternatives.
WHO SHOULD SKIP IT
Buyers who need precise control over inflection and prosody, non-English or mobile-only workflows, or anyone uncomfortable with their voice data being used to improve Descript's product should look elsewhere.
Synthesised from 52 real owner comments across 2 platforms. Every point is grounded in the comments — no marketing, no AI guessing. How we do it →