If auto captions keep spelling names wrong in your videos, the cause is simple. A speech engine writes what it has heard before, and it has heard common words far more often than your surname, your company or the village your guest comes from. So it guesses. Because the guess comes from the same model, it is the same wrong guess every time. You fix it on Monday and fix it again on Thursday.
Titro’s personal dictionary closes that loop. It looks at the corrections you make in the editor, notices when you make the same fix in two different projects, and from then on writes the word your way in every new transcript. Here is exactly how it decides, and how to steer it.
Why names are the words that go wrong
Everyday words are easy for speech recognition because they turn up constantly in the speech it learned from. Names, places and brands are the opposite. They are rare, often borrowed from another language, and spelled in ways the sound doesn’t give away. The audio alone can’t tell the engine whether your guest is Jon, John or Gjon.
They are also the words viewers notice. Nobody spots a misspelled “the”. A misspelled guest name on screen looks careless, and the guest notices first.
How the personal dictionary learns
Every time you save an edit, Titro compares the captions before and after and picks out the words you replaced. One fix is only noted. When you make the same fix in a second, different project, the word is learned. Every transcript after that already carries your spelling before you open the editor.
| What happens | |
|---|---|
| First video | The engine writes “Prishtin”. You correct it to “Prishtinë”. Titro notes the fix. |
| Second video | Same mistake, same fix, in a different project. The word is learned. |
| Every video after | The transcript already says “Prishtinë”. |
It counts projects on purpose. Fixing the same name in three captions of one interview counts once, because it is one video and one decision. A second video shows a habit, and a habit is worth a rule.
What it won’t learn
A feature that rewrites what you said has to be careful, so the rules are strict:
- One word for one word. If you rewrite two words into three, that is an edit to the sentence, and nothing is learned from it.
- Spelling fixes only. The new word has to look like the old one. Replacing “said” with “declared” changes the wording, so it never becomes a rule.
- At least three letters, and no numbers. A rule that turned every “a” into “e” would wreck every transcript you ever made.
- No case endings. Albanian changes the end of a word with its role in the sentence, so a fix from “Prishtinë” to “Prishtinës” depends on that sentence and is never learned as a rule.
- The newest answer wins. If you later correct the same word to a different spelling, the new one replaces the old and the count starts again.
Capital letters are handled for you. A name you wrote with capitals keeps them wherever it lands, and a lower-case word gets a capital when the original had one, at the start of a caption for example.
Source language only
The dictionary works on the language that was spoken. It rewrites the words the engine heard and learns from edits to those captions. On a translated project, edits to the translated lines teach it nothing, because comparing Albanian words with English captions would produce nonsense rules like “universiteti” becoming “university”. A translation is never rewritten by the dictionary either.
Private to your account
The list belongs to your account. It never leaves it and is never exported. It also has its own switch, separate from Help Titro get better, which decides whether your corrections and that clip’s audio may be kept to improve the speech recognition. You can turn the research switch off and keep the dictionary on, so your own name is still spelled right.
Turning the dictionary off deletes the list, so no old rule keeps rewriting your transcripts in the background. Deleting your account deletes it too.
How to check and edit your list
- 1
Open your profile
In the iPhone or Android app, go to Profile. The switch is called Remember the words I correct, and it is on for new accounts.
- 2
Read the list
Each entry shows the wrong spelling, an arrow and your spelling. A word you have fixed in only one project so far is marked not yet.
- 3
Add a word by hand
Type the wrong word and how it should be, then tap the plus. A word you add yourself applies straight away, without waiting for a second project. One word on each side, at least three letters.
- 4
Remove an entry
Tap the cross next to it. Future transcripts go back to what the engine hears.
The list itself lives in the app and holds up to 500 words. Corrections you make in Titro Studio on the web count the same way, because the learning happens on the server for your account.
When the dictionary isn’t enough
The dictionary fixes words the engine gets wrong in a consistent way. For a video full of names you have never said on camera before, the teleprompter is quicker. You write the script, read it while you record, and the subtitles follow what you actually said, with the names spelled the way your script has them.
Fix a name in two videos and you won’t type it again. More on how Titro writes the transcript in the first place is on the auto captions page. Your first video is free, up to three minutes.
