Improving speech recognition for specific terms

Learn how to improve speech-to-text accuracy by adding terms to your personal speech recognition dictionary in Rad AI Reporting.

In this article:

1. Prerequisites
2. How it works
3. Adding a term to your dictionary
   1. From the report editor
   2. From Personal Settings
4. Managing dictionary terms
   1. Switching between dictionary scopes
   2. Editing or removing a term
5. Term requirements
6. Tips and best practices
7. Frequently asked questions
   
    

Prerequisites

Before you start, make sure you’re using the Advanced speech service. Other speech services do not support the speech recognition dictionary. 

 

How it works

Adding terms to your speech recognition dictionary helps the system recognize the words and phrases you dictate most often. When a term is added, the speech recognition engine is more likely to choose that term over similar-sounding words during transcription.

This can help in two ways:

• Improved recognition of existing terms — Increase accuracy for words the system already knows but may transcribe incorrectly, such as uncommon medication names or specialized medical terminology.
   
• Recognition of new terms — Teach the system words or phrases it may not recognize at all, such as physician names, facility names, study protocol names, or other organization-specific terminology.

⚠️ Because added terms are prioritized during transcription, they can sometimes affect recognition of similar-sounding words. For best results, only add terms you expect to dictate regularly.

The speech recognition dictionary exists in two scopes:

• Personal — Terms are managed by individual users and speech recognition improvements apply to the individual user only. 
   

• Organization —  Terms are managed by Rad AI and speech recognition improvements apply to all users at your organization. Terms in the organization dictionary are visible (read-only) to individual users. Learn how to switch dictionary scopes.

  💡 Contact [e-mail] to request the addition of specific terms to your organization dictionary.

If the same term exists in multiple scopes, the personal term takes precedence over the organization term.

Rad AI may apply additional speech recognition improvements behind the scenes.

 

Adding a term to your dictionary

From the report editor

You can add a term to your speech recognition dictionary directly from the report editor while working on a report.

1. In the report editor, with a report open, highlight the word or phrase you want the engine to recognize better.
   
   
   
    
2. Right-click the selected text, then click Improve speech recognition.
   
   
   
    

3. In the dialog that appears, review the term and edit if needed. 

   ⚠️ An error will show if the term doesn’t meet the requirements.
   💡 The system applies the same preference to the singular or plural form of the term when applicable. For example, adding “biopsies” also improves recognition of “biopsy”.
4. Click Add to dictionary. The term will appear in your dictionary for improved speech recognition.
   
   
    

 

From Personal Settings

1. Click on your initials at the bottom of the main navigation menu, then select Personal Settings.
   
   
   
    
2. Click or scroll down to Speech > Speech Recognition Dictionary.
   
   
    

3. Place your cursor in the Add a term… field and enter the term you want to add. 
   
   
   
    

4. Click Add, or press Enter. The term will appear in your dictionary for improved speech recognition.

   ⚠️ An error will show if the term doesn’t meet the requirements.
   💡 The system applies the same preference to the singular or plural form of the term when applicable. For example, adding “biopsies” also improves recognition of “biopsy”.

Managing dictionary terms

Switching between dictionary scopes

Use the Scope selector to switch between your personal terms and your organization’s terms. Your personal terms are editable, while organization terms are read-only.

💡 Both lists are combined when speech is sent to the engine, so organization terms still apply to your dictation even though you can’t edit them here.

1. Go to Personal Settings > Speech > Speech Recognition Dictionary.
    
2. Click the Scope drop-down and select either Personal or Organization.
   
   

 

Editing or removing a term

1. Go to Personal Settings > Speech > Speech Recognition Dictionary.
    
2. Do one of the following:
    
   • Scroll down the list and locate the term you want to edit. 
      
   • Click Search terms and enter the term you’re searching for. The list will update automatically as you type, hiding any terms that don’t match your search. 
     
      
3. Then do one of the following: 
    
   • To edit the term, click Edit, modify the term, then click Save.
      
   • To remove the term from your dictionary, click Remove. The term will be deleted instantly, and speech recognition improvements will no longer apply.
      

 

Term requirements

You won’t be able to add a term to your dictionary if:

• The term is shorter than 4 characters. 
• You’ve reached your personal dictionary limit of 50 terms
• The term already exists in your personal dictionary or the organization dictionary.
• The term is on Rad AI’s list of common English and medical terms. This is because prioritizing them could reduce transcription accuracy for similar words or phrases.

 

Tips and best practices

• Add only terms you dictate regularly. Each term in the dictionary nudges the speech recognition engine away from similar-sounding words.
   
• Add the form of the term you actually say most often. The system supports singular and plural variations of any term, when applicable. 
   
• If a short abbreviation is rejected because it’s under 4 characters, try adding a longer phrase that includes it instead. For example, use “Macro MRB” instead of just “MRB”.
   
• If a term is rejected as a common word, it usually means the engine already models it well and adding it to your speech recognition dictionary would do more harm than good.
   
• If a term already appears in your organization’s dictionary, you don’t need to add it to your personal dictionary because it already applies to your dictation.

 

Frequently asked questions

What’s the difference between personal and organization terms?

Personal terms are managed by you and can be edited in Personal Settings. 

Organization terms are managed centrally and appear as read-only in Personal Settings—but they still apply when you dictate.

 

Why can’t I see the speech recognition dictionary in Personal Settings?

Make sure you meet the Prerequisites for using this feature.

 

When do added or removed terms take effect?

Changes take effect the next time dictation starts.

 

Why can’t I add a short term or abbreviation?

Terms must be at least 4 characters long. If you want to support a short abbreviation, use a longer phrase that includes it—as long as it matches how you say it during dictation.

 

Is a speech recognition improvement applied to both singular and plural variations of the term?

Yes, where applicable. The only exception is certain terms on Rad AI’s list of common English and medical terms, to avoid degrading accuracy. 

 

What happens with capitalization and casing when dictating terms from my dictionary? 

Dictated terms generally try to follow the casing you entered, but normal report formatting rules still apply. For example, if you add "biopsy" to your dictionary and dictate it right after a period—as in "Findings included below."—it may appear as "Findings included below. Biopsy" because the report editor still applies capitalization after punctuation

 

What’s the difference between the speech recognition dictionary and substitutions?

Use the speech recognition dictionary when you want Rad AI Reporting to better recognize a word or phrase as you dictate. It works during speech recognition, increasing the likelihood that the correct term is transcribed, including similar-sounding words or phrases.

Use substitutions when you want specific dictated text to be replaced with different text. Substitutions only work if the speech engine first returns the word or phrase you set up, so they’re best for direct, predictable replacements such as spoken shortcuts that expand into longer text.