Fix a robotic or unnatural-sounding voice
If your dubbed voice sounds robotic, flat, or off-tone, the issue is almost always fixable from your side — usually without re-uploading the source. Here's what to try, in order of biggest impact.
1. Fix the transcript
The voice reads exactly what's in the transcript. Misheard words, missing punctuation, and merged sentences all push the voice into unnatural cadence.
Check for:
Misheard words — especially technical terms, brand names, proper nouns
Missing or merged sentences
Punctuation — commas and periods control rhythm and pauses
What to do:
Open the transcript editor, correct the affected segments, and click Save
Only the edited segments re-render — no full-video reprocess needed
Even small punctuation fixes can change how a sentence flows
2. Set the right pronunciation for names, brands, and acronyms
Brand names, product names, and acronyms are common sources of robotic-sounding output. Instead of fighting it segment-by-segment, fix it once at the Translation Style level.
Please be aware that changes made to the Translation Style are not retroactive. These settings will only apply to future videos you upload. Existing videos or projects already in progress will not be updated automatically with these changes.
What to do:
Open your Translation Style and go to the Pronunciations tab
Add an entry for each name, brand, or acronym with how it should be read
This is the right place for:
Company and product names
Industry-specific terminology
Foreign words used inside another language
Acronyms that should be spelled out (or kept as a single word)
For a full walk-through, see Translation Styles.
3. Improve your source audio (for the next dub)
The AI can only sound as good as the audio it receives. The fixes above will rescue most already-rendered dubs — but for future videos, cleaner source = more natural voice from the start.
Aim for:
Clean audio — minimal background noise, music, or echo
Clear pronunciation — mumbled or rushed speech produces unreliable transcripts, and the voice inherits those errors
No heavy effects — reverb, auto-tune, or compression confuse the voice model, especially when cloning
If you can only work with existing footage, isolate the vocal track before uploading.
4. Re-generate a glitchy segment
Sometimes a single segment comes out with an audible artifact — a stretched-out vowel like "oooooooh", a stutter, a clipped or doubled word, garbled syllables, or a brief stretch of nonsense audio. These are almost always one-off TTS hiccups, not a problem with your source or voice choice. Re-generating just that segment usually produces clean audio on the next pass.
What to do:
Open the Advanced Editor and find the affected segment
Make a small edit to trigger a re-render — retype a word, toggle a punctuation mark, or delete and re-add a single character
Click Save — only that segment's audio is re-generated, no extra credits charged
Listen to confirm. If the glitch is still there, save the segment again — a second TTS pass almost always lands cleanly
If the same glitch keeps coming back on the same segment after two or three regenerations, or if the issue affects many segments across the video, it's not a one-off — check the earlier sections (transcript, pronunciations, source audio) or contact support.
When to contact support
If the voice still sounds wrong after the steps above, write to hello@dubly.ai with:
The dub link
The exact segment(s) where the voice sounds wrong
Which voice you're currently using
A short reference clip of how you'd want it to sound, if you have one