Dictionary
Improve transcription accuracy with custom vocabulary.
Custom dictionaries work at the speech recognition level, helping transcription models correctly recognize and spell domain-specific terms, names, brands, and technical jargon. This feature is supported by GPT-4o Transcribe, GPT-4o mini Transcribe, Soniox Realtime, and Soniox Async models.
For simple find-and-replace rules, see Word Replacements.
How to Access
Navigate to Dictation Models in the settings, then click the Settings button next to a supported model.

Setting Up Your Dictionary
The model settings dialog contains a Prompt (Dictionary) field. Enter context or specific terms you want the model to recognize accurately.

Tips for Effective Dictionaries
- Include proper nouns: Names of people, companies, products, and places that might be uncommon.
- Add technical terms: Industry-specific jargon, acronyms, or specialized vocabulary.
- Provide context: A brief sentence describing what you'll be discussing helps the model understand the domain.
- Use correct spelling: The model will follow the exact spelling and capitalization you provide.
Example Prompts
For a product meeting:
The meeting covers GPT-4.5 updates, so transcribe any mentions clearly.For a medical dictation:
Medical terms: amoxicillin, hypertension, systolic, diastolic, lipid panel.For a team standup:
Names: Priya, Kenji, Omar. Projects: ProjectAlpha, Sprint-42, CI/CD pipeline.Supported Models
| Model | Dictionary Support |
|---|---|
| GPT-4o mini Transcribe | Yes |
| GPT-4o Transcribe | Yes |
| Soniox Realtime | Yes |
| Soniox Async | Yes |
| Whisper models | Use Whisper Prompting instead |