Fix errors in your translation
If your translation has missing words, wrong names, or misheard phrases, here's how to fix them — and how to prevent the same errors next time. For a general walkthrough of the editor, see Reviewing and Editing Transcriptions.
Quick fix: one-off errors
Open the dub, go to the Edit Translation tab, click the sequence with the error, fix the source text, and save. Dubly re-translates the sentence and regenerates the dubbed audio automatically. Use this for typos, wrong numbers, and single-word mistakes.
Fixing common error types
Misheard brand, product, or person names
Example: "Dubly" transcribed as "Doubly" or "Anthropic" transcribed as "Anthrophic".
Fix the sequence in the editor.
Also add the term to your Translation Style (Translation Style → Do-Not-Translate list, or as a Custom Word Pair with the exact spelling). Future dubs pick it up automatically — no more per-sequence fixes.
Numbers and dates
Example: "twenty-twenty-six" transcribed as "2026" or vice versa.
Fix the source text. If the AI is struggling with pronunciation, the most effective fix is to spell the number out as text (e.g., change "2026" to "twenty twenty-six"). This forces the AI to follow the exact phonetic structure you want.
If a specific format matters for your brand (currency, date style), add it as a custom word pair in your Translation Style for the target language.
Technical terms and acronyms
Example: "API" pronounced as "apee" in the dubbed audio.
Fix the source text if transcribed wrongly.
Add a pronunciation hint in your Translation Style so the voice model reads it correctly next time.
Missing or merged sentences
If two sentences were merged into one sequence, or a short sentence is missing, you can edit the text in that sequence to match what was actually said. Dubly can't split one sequence into two — if a merge is making a line unusable, edit the text to be coherent within the sequence's timing.
Wrong speaker assignment
If the diarizer labeled the wrong person as the speaker:
In the editor, change the speaker for the affected sequence.
The dubbed audio regenerates with that speaker's voice.
Long stretches of silence transcribed as filler words
Sometimes background noise is translated as "um", "uh", or meaningless phrases.
Delete the content of the source text, then save.
Preventing errors next time
Most translation errors come from the source audio. Before uploading:
Record clean voice audio — quiet room, decent microphone, minimal background music under dialogue.
Avoid heavy voice effects (deep reverb, telephone filter) — they throw off the speech recognizer.
Set the correct speaker count. A video with two speakers but "1 speaker" set will merge the second speaker's lines into the first — the translation becomes unusable for that speaker.
Build out your Translation Style with your brand's vocabulary. Every term you add is one fewer error to fix manually.
Still wrong after fixing?
If a term keeps coming out wrong even after you've added it to your Translation Style, contact hello@dubly.ai with the dub title, the source sentence, and what you expected.