AutoCorrect+ (Mark & Fix)
Mark transcription mistakes by category and resolve them manually, with AI, or dismiss them.
Even great transcription occasionally mishears a name, a piece of jargon, or a fast phrase. AutoCorrect+ is Kalima's two-step "Mark & Fix" workflow: you flag the exact words that bother you, then resolve each one on your own terms by typing the fix yourself, asking AI for a suggestion, or dismissing it. Corrections are applied word by word, so your playback timing and karaoke highlighting stay perfectly in sync.
How Mark & Fix works
AutoCorrect+ separates spotting an error from fixing it. As you read or listen to a transcript, you mark anything that looks wrong and tag it with a category. Each mark becomes an item in a resolve queue you can work through whenever you like, one at a time or in a quick cleanup pass at the end.
The workflow has three stages. First, Mark a word or phrase by selecting it, right-clicking, and tagging it with a category. Next, Resolve the mark by typing the correction, asking AI for a suggestion, or dismissing it. Finally, you can optionally Apply everywhere to fix every other occurrence, where AI confirms each one is the same mistake first.
Because the actual edit happens at the word level, fixing a transcript never breaks the audio sync: click-a-word-to-jump and the word-by-word karaoke highlight keep working exactly as before.
Marks are anchored to the specific words you flag and survive page reloads, so you can mark issues during a live session and resolve them later. Marking is available on saved sessions, so start the session and let Kalima save it, then mark and fix.
Step 1: Mark a mistake
The six mark categories
Tagging each mark tells AutoCorrect+ (and you) what kind of fix is needed, and lets you filter the queue later.
| Category | Use it for |
|---|---|
| Wrong word | A word that was misheard or garbled. |
| Name / term | A misheard person, company, product, or piece of jargon. |
| Grammar | A grammatical slip in the transcribed text. |
| Punctuation | Missing or incorrect punctuation. |
| Filler / remove | Filler words to delete (for example "um", "uh"). |
| Note | A reminder or annotation for yourself, with no replacement needed. |
Step 2: Resolve each mark
Open the AutoCorrect+ (Mark & Fix) panel to see every open mark. Work through them with whichever approach fits, and you can mix approaches across the queue.
Type the correct text into the mark and Apply. This is a straightforward local edit that doesn't use any of your AI allowance. Use it when you already know exactly what was said, or for the Filler / remove category, which deletes the marked text instead of replacing it.
Every mark moves through three statuses so you can always see where you stand:
- Open: flagged but not yet handled.
- Resolved: a correction was applied.
- Dismissed: set aside without a change.
Made a fix you didn't mean to? Use Undo to revert the last applied correction. The mark returns to your queue so you can handle it differently.
Filter the queue
If you've collected a lot of marks, use the category filter chips at the top of the panel to focus on just one type at a time. For example, clear all your Filler / remove marks in one pass, then move on to Name / term fixes.
Step 3: Fix the same mistake everywhere
When the same word is misheard repeatedly, such as a recurring name, a product, or a piece of jargon, you don't have to mark every instance.
Apply everywhere is context-aware on purpose. If the same spelling is actually correct somewhere else in the transcript, AI leaves that instance alone, so you won't accidentally "correct" a word that was right. Each occurrence it validates counts toward your AI autocorrections allowance.
What it costs against your AI allowance
AutoCorrect+ is part of Kalima's AI toolset, so the AI-powered actions are metered by your plan's weekly AI autocorrections quota, while manual edits are free.
| Action | Uses AI quota? |
|---|---|
| Typing a correction yourself | No, it's a local edit |
| Removing a filler word manually | No |
| Asking AI for a suggested fix (the sparkle) | Yes |
| Apply everywhere (validating each other occurrence) | Yes, one check per occurrence |
| Dismissing a mark | No |
AI-generated text, including AutoCorrect+ reasoning, is written in your account's display language. See AI Usage & Limits for the per-plan numbers and what happens when you reach a limit.
Tips & limitations
- Names everywhere, fixed once. Mark a misheard name, resolve it, and tick also fix others to clean it up across the whole transcript in a single pass.
- Tidy up filler. The Filler / remove category deletes the marked words rather than replacing them, which is ideal for stripping "um" and "uh".
- Fix the translation by fixing the source. When a translation reads oddly because the original was misheard, correcting the original word improves what's there to translate.
- Minimal by design. AI fixes correct mistakes but won't rewrite or pad your transcript with words nobody said.
- Marks that can't be placed are set aside. If the marked text can no longer be uniquely located in its segment, for example because the surrounding transcript changed, AutoCorrect+ dismisses that mark rather than guessing. Re-mark the current text and apply the fix again.
- A short cool-down on bursts. All AI features share a brief per-user rate limit, so firing many AI suggestions in quick succession may ask you to wait a moment before trying again.
Frequently asked questions
No. AutoCorrect+ applies corrections at the word level, so playback timing, click-a-word-to-jump, and the word-by-word karaoke highlight all stay aligned after a fix.
Mark one occurrence, type or AI-suggest the correction, and tick also fix others before applying. AI reviews each remaining occurrence and replaces only the ones that are genuinely the same mistake.
That's intentional. AI deliberately skips any occurrence where the original word was actually correct in that context, so it never "corrects" something that was right. If any remaining ones really are errors, fix them individually.
The marked text could no longer be uniquely located in its segment, usually because the surrounding transcript changed since you marked it, so AutoCorrect+ set the mark aside instead of guessing. Re-select the current text, mark it again, and apply your correction.
No. Typing a fix yourself (or removing a filler word manually) is a plain edit and doesn't count against any AI quota. Only the AI suggestion (the sparkle) and the per-occurrence checks during Apply everywhere use your AI autocorrections allowance.
Yes. Use Undo to revert the last applied correction; the mark returns to your queue so you can resolve it differently or dismiss it.