Custom vocabulary and text replacements

TranscribeX has two tools for words the model gets wrong. Both are in the Dictionary section of Settings.

• Custom Vocabulary helps the model recognize names and specialized terms while it transcribes.
• Text Replacements rewrite the text afterwards, swapping one word or phrase for another every time it appears.

Use Custom Vocabulary for things the model mishears in different ways. Use Text Replacements when the same mistake shows up the same way every time.

Custom Vocabulary

Custom Vocabulary works with English audio on Parakeet V2, Parakeet V3 and the Whisper models. Other models ignore it.

1. Open Settings › Custom Vocabulary.
2. Turn on Enable Custom Vocabulary.
3. In Custom Vocabulary Words, type the terms you need: product names, people’s names, technical words. Separate them with spaces or commas.
4. Transcribe a file as usual.

It applies when the transcription language is English or Auto Detect. With Auto Detect, the words are only used if the audio turns out to be English. While it’s working, the progress label shows “Preparing recognition vocabulary...”. The first time, TranscribeX downloads a small extra English model for this step.

Tips:

• Enter each word the way you want it spelled, including capitals (for example TranscribeX, Kubernetes).
• Duplicates are ignored, whatever their capitalization. The first spelling you type is kept.
• Because spaces separate entries, a multi-word name is treated as separate words.
• Keep the list focused. A short list of the terms you actually hear works better than a long dictionary.

If vocabulary preparation fails, TranscribeX carries on and transcribes without it.

Text Replacements

Text Replacements find a word or phrase in the transcript and replace it. They run on every model when a file or recording is transcribed, and again when you reflow a transcript.

1. Open Settings › Text Replacements.
2. Make sure Enable Dictionary Replacements is on (it is by default).
3. Click Add Entry and fill in the new row.

Each row has these columns:

The checkboxes in the Case and Enabled headers toggle that column for every row. Use the Search field to find a rule in a long list.

Note: Replacements match any part of the text, not only whole words. A rule from cat to dog also turns category into dogegory. Make the Source specific enough to avoid this. Rules run in the order you created them.

Replacements apply to new transcriptions. To fix a transcript you already have, transcribe it again or use search and replace in the transcript view (see editing transcripts).

Import and export replacements as CSV

Use Import and Export in the Text Replacements pane to back up your rules, move them to another Mac, or edit many rules at once in a spreadsheet.

The file is a UTF-8 CSV with a header row first. TranscribeX skips the first line when importing, so always include it. Each data row has four columns:

1. Source
2. Target
3. Case Sensitive: 1 or 0
4. Enabled: 1 or 0

Source,Target,Case Sensitive,Enabled
transcribe x,TranscribeX,0,1
mac whisper,MacWhisper,0,1
Acme,ACME Corp,1,1

Rows with fewer than four columns are skipped. Import adds the rows to your existing rules and doesn’t replace them, so importing the same file twice creates duplicates. Export saves all your rules to dictionary.csv by default.