Vocabulary and replacements

The speech model knows everyday language well and your own names and terms less well. Dictate gives you two tools for that in Settings under the Speech group. The vocabulary field tells the model what to expect. The replacements table turns a spoken phrase into the exact text you want typed. Two toggles clean up the result. This guide explains each one and the order they run in.

The vocabulary field

The vocabulary is a free text field of up to 2000 characters. Write the names and terms and abbreviations you dictate often. Separate them with spaces or line breaks. Dictate hands the field to the model as a hint before every transcription so a name it would otherwise guess at comes out right.

The vocabulary shapes what the model hears. It does not replace anything. For an exact replacement use the table below.

The replacements table

A replacement is one row with a spoken phrase and the typed text. When the transcription holds the spoken phrase Dictate replaces it with the typed text before the text is inserted. The table holds up to 200 rows and has an Add button and a Remove button for the selected row.

The classic use is an email address. Say my email and type name@example.com. Dictate replaces the two words with the address every time.

The rules of a match are these. The spoken phrase is matched case insensitively so My Email and my email both match. It matches whole words only so a phrase inside a longer word is left alone. Longer phrases are tried before shorter ones so a row for my work email wins over a row for my email when both are present. The typed text is inserted exactly as you wrote it with its case and punctuation.

A spoken phrase holds 1 to 100 characters and the typed text 1 to 500 characters. An empty row is not saved.

Remove filler words

The Remove filler words toggle is off by default. Switched on it removes the words um and uh and er wherever they stand alone as words together with the comma or period the model wrote after them. The word like is removed only when the model set it off with commas on both sides or when it opens the text and is followed by a comma. So I like this stays as it is and like I said becomes I said.

Capitalize and end with a period

The Capitalize first letter and end with period toggle is on by default. Switched on it uppercases the first letter of the text when the text starts with a letter and appends a period when the text does not already end in a period or a question mark or an exclamation mark or a colon or a semicolon. A closing quote or bracket after such a mark counts as an ending too so a quoted sentence is left alone.

Switch it off when you dictate fragments into a form or a spreadsheet or the middle of a sentence.

The order of the steps

Every transcription runs through the same steps in this order. Dictate trims the text and collapses runs of spaces into one. Then it applies the replacements. Then it removes the filler words when the toggle is on. Then it capitalizes and adds the period when the toggle is on. Then it decides about the leading space described in the Hotkeys and modes guide. A text that is empty after the first step inserts nothing and writes no row to History.

Examples

You say my email is my email and the row my email to name@example.com is present. The text reads My email is name@example.com.

You say um I think we should like move the meeting with a pause around like and the filler toggle on. The model writes the pauses as commas and Dictate removes um and the like between the commas. The text reads I think we should move the meeting.