Guides
Reading live transcript corrections
Once a line of live dictation settles, SpeakWith can mark it in two ways. Inside the line, color and underlines show which words you changed yourself. After the line, small badges show what SpeakWith corrected on its own and, if you have switched them on, what its extra checks found. Most lines carry no marks at all. The diagram below shows one line with every kind of mark on it.
Marks inside the line
The line you see is not always the line you spoke. You can open any settled line and fix it with the keyboard, or dictate a phrase straight into it, and the day document keeps the result. Playback, though, still plays the original audio. So each word is painted by where it came from, and the words the recognizer wrote stay plain.
| Mark | What it means |
|---|---|
| Typed words | Your typed-words color, over a pale band of the same hue. If you said three headshots and the line came back as tree headshots, opening it and typing three marks only three. |
| Dictated into a line | The text keeps its normal color, sits on a pale band of the spoken-into-line color, and takes a dotted underline in a complementary shade. Open a line, speak a phrase, and the phrase lands marked this way. |
| Deleted words | Still visible, struck through and faded, like tracked changes in a word processor. A line you removed entirely is struck through as a whole and dimmed. |
| Plain words | The recognizer wrote them and nobody changed them. No color, no band. |
Both colors are yours to choose under Settings, Playback: Typed words and Spoken-into-line. The band and the dotted underline are derived from the color you pick, so the two families stay distinct at a glance.
Only your own edits are painted. When SpeakWith joins two spans into one sentence, or a second engine re-reads a line, those words were all spoken, so they stay plain.
Badges after the line: corrections SpeakWith made
A badge trails the text of the line it belongs to. There is no reserved column, so a line with nothing to report looks exactly like every other line. When several corrections land on one line, they share a badge, separated by commas. Hover over any badge to read a full explanation.
| Badge | Meaning |
|---|---|
↦ alia → Aaliyah | A rewrite from your Special Terms list. The first arrow marks the list, and the second reads heard, then written: the recognizer wrote alia, and your rule for the name Aaliyah replaced it. This layer is always on: the badge appears whenever a listed term was rewritten, whatever mode the recorder is in. |
Akheffer → Okafor | The term spotter heard one of your special terms in the audio and corrected the line to match. The arrow means the text changed. |
Nguyen | The spotter heard a term but the line was not changed: usually the term was already spelled correctly, or correcting is switched off. The tooltip says which. |
⚡ Akheffer → Okafor | The match came from one of your Voice Examples, a recording of you saying the name, rather than from its spelling. The bolt is reserved for that. Okafor is the kind of name that earns one: no spelling rule covers how it is misheard. |
The spotter is set up under Settings, Corrections. Show spotted special terms hangs the badge on every line where a term was heard. Correct spotted special terms in the line lets the spotter change the text, and a badge with an arrow appears whenever it does, so a correction never lands silently.
An optional second opinion
One more check can run on each line. It is off by default and lives under Settings, Debug, because it is for understanding the engines rather than for everyday dictation. When it is on, one more glyph trails the line.
The second pass re-reads short lines with a second engine and reports with a single character.
| Glyph | Meaning |
|---|---|
… | Queued. The second engine reads the line again once the take stops. |
✓ | Green. The second engine heard the same words. |
≠ | The second engine heard something different. A row appears beneath the line with the other reading, so you can compare the two. |
? | Orange. The second engine returned nothing, or could not read the audio. Its model may be missing or still loading. |
The switch is Show second-pass diagnostics. It never changes what is written. It only shows what happened.
When nothing appears
An unmarked line is the normal case, and it can mean several things. The recognizer got the line right and nobody changed it. A special term was already spelled correctly, so there was nothing to correct. Or the second pass is simply switched off, so its glyph draws nothing.
If you expected a correction and see no badge, check the two spotter switches under Settings, Corrections first, then that the term is in your Special Terms list or has a Voice Example.
See also: reading the recorder badge, which reports your correction rate over time.