My voiceover sounds inconsistent or distorted
Sometimes a finished dub plays through fine on most segments but sounds wrong on others — distorted, robotic, suddenly louder or quieter, or like a different speaker mid-sentence. Here's how to track down what's happening and fix it.
Common causes
The source audio is the real problem
Voice cloning is only as good as the source. If your input has:
background music or strong sound effects under the speech,
multiple speakers overlapping,
echo, reverb, or room noise,
low recording volume with normalized highs,
…the cloned voice will sound inconsistent. The clone is matching what it heard, including the noise.
What to do: Improve the source recording. Better mic placement, less reverb, and isolated speech (no music underneath) make a huge difference. If you can't re-record, export the source with cleaned-up audio before uploading.
Speaker detection misassigned a segment
When more than one speaker is on camera, the system splits segments by speaker. If a segment was assigned to the wrong speaker, that segment will sound like a different voice.
What to do: Open the segment, check the speaker assignment, and reassign it to the correct speaker. The audio regenerates automatically.
The voice clone itself was weak
Sometimes the cloned voice just doesn't capture the speaker well — usually because there wasn't enough clean speech in the source to clone from.
What to do: Switch the segment (or the whole dub) to a Studio voice instead of cloning. Studio voices are professionally trained and consistent across every segment.
One bad segment
A single segment can come out distorted even when the rest is fine, especially when the source audio for that segment was rough (background bang, music swell, overlap with another voice).
What to do: Open the segment and trigger a regeneration. If it's still bad, edit the segment text — sometimes a small wording change is enough to produce a clean take.
The voice suddenly whispers, changes accent, or breaks mid-segment
These are signs that the voice generation hit a tricky spot. The model usually self-recovers within a few words, but on harder segments it can:
Drop into a whisper — usually triggered by very quiet input audio in the source segment, or by long pauses the model interprets as soft speech.
Shift accent — the cloned voice may briefly slip into a different regional accent if the source recording had inconsistent accent across segments.
Break or stutter — the model loses confidence and produces audio glitches, often around long words, technical terms, or numbers.
What to do: Open the affected segment and edit either the original or the translated text slightly (rephrase, simplify, or add punctuation to give the model a clearer parse), then save. The audio regenerates with the new text — usually cleanly. If it keeps happening on the same segment, switch that segment to a Studio voice.
The voice is monotonous, too chaotic, or doesn't sound like the speaker
These are tone and emotion issues, not technical glitches:
Monotonous — the dubbed audio is flat, no rise and fall, no emotion. Usually happens when the source speech is itself flat (recorded reading from a script with no expression).
Too chaotic — the dub has too much emotional variation, sounds excited or anxious where the source was calm. Rare, but happens with very dramatic source recordings.
Doesn't sound like the speaker — the cloned voice doesn't match the source person well enough.
What to do:
For monotonous output: improve the source recording. Coach the speaker to add intentional pacing and emphasis. The clone matches whatever it heard.
For doesn't-sound-similar: the source audio probably wasn't long enough or clean enough to clone from. Use a longer, cleaner source segment, or fall back to a Studio voice.
For too chaotic: usually self-corrects on regeneration. If it keeps happening, switch the segment to a Studio voice for that line.
What to do before contacting support
Identify the specific segment(s) that sound wrong.
Listen to the original audio for those segments — does the source itself sound rough?
Try the fixes above.
When to contact support
If most of the dub is bad rather than just a few segments, or if you've already tried regenerating the affected segments and they still come out wrong, contact support with the dub URL and a list of which timestamps sound off.