Skip to main content

Using Custom Models

Learn how to create, train, and manage custom language models in Scriptix to improve transcription accuracy for your specific domain.

What are Custom Models?​

Custom models are specialized speech recognition models that you can train with your own audio data to improve accuracy for:

  • Domain-specific terminology
  • Industry-specific vocabulary
  • Particular speakers or accents
  • Specific audio environments

Accessing Custom Models​

  1. Navigate to the Custom Models page in Scriptix
  2. View all your organization's custom models in the data grid

Understanding the Custom Models Page​

Models List​

The data grid displays your custom models with the following information:

Columns:

  • Name - Custom model name
  • Training Status - Current status with color-coded badge:
    • Gray (Not Running) - Model created but not started training
    • Amber (Ready to Run) - Ready to begin training
    • Blue (Running) - Currently training
    • Green (Success) - Training completed successfully
    • Red (Failed) - Training failed, needs attention
  • Base Language - Language the model is trained for
  • Last Modified - Most recent update date

Actions:

  • Edit (pencil icon) - Open model details to manage training
  • Delete (trash icon) - Remove the custom model

Creating a Custom Model​

Step 1: Access Create Modal​

  1. Click the "Create Custom Model" button (Plus icon) in the top right
  2. Create Custom Model modal opens

Step 2: Configure Model​

Model Name (Required):

  • Enter a descriptive name for your model
  • Example: "Medical Cardiology EN", "Legal Depositions NL"
  • Helps identify the model's purpose

Base Language (Required):

  • Select from dropdown of trainable languages
  • Choose the language of your training audio
  • Only languages supported for custom model training appear

Step 3: Create​

  1. Review the entered information
  2. Click "Create" button
  3. Model is created with "Not Running" status
  4. Click on the model to access training interface

Getting Started Guide​

When you first visit the Custom Models page, a guide modal appears automatically with information about:

  • What custom models are
  • How they work
  • Steps to create and train

Click "Getting Started" button anytime to view the guide again.

Training Your Custom Model​

After creating a model, you'll need to train it with your audio data:

  1. Click Edit (pencil icon) or click on the model name
  2. Training interface opens (model detail page)
  3. Follow the training workflow to upload training data and start training

Note: Training details and data management are handled in the model detail page.

Managing Custom Models​

Edit/View Model Details​

  1. Click the Edit icon (pencil) or model name
  2. Opens model detail page where you can:
    • View training status
    • Upload training data
    • Start/manage training
    • View model configuration

Delete Custom Model​

  1. Click the Delete icon (trash) on any model
  2. Confirmation modal appears
  3. Review model name to confirm deletion
  4. Click "Delete" or "Confirm"
  5. Model is permanently removed

Warning: Deleting a custom model is permanent and cannot be undone. Ensure you no longer need the model before deleting.

Training Status Explained​

Not Running (Gray)​

  • Model has been created
  • No training data uploaded yet
  • No training started
  • Next step: Upload training data

Ready to Run (Amber)​

  • Training data has been uploaded
  • Model is ready to begin training
  • Next step: Start the training process

Running (Blue)​

  • Training is currently in progress
  • System is processing your training data
  • Can take several hours depending on data size
  • Wait for: Training to complete

Success (Green)​

  • Training completed successfully
  • Model is ready to use
  • Can be selected for transcription
  • Next step: Use model in transcriptions

Failed (Red)​

  • Training encountered an error
  • Check error messages for details
  • May need to review training data
  • Next step: Check logs, fix issues, retry

Using Your Trained Model​

Once training status shows "Success":

  • Model is available for selection when creating transcriptions
  • Select your custom model during file upload
  • Improved accuracy for your specific domain
  • Can be used in batch transcription or real-time

Best Practices​

Naming Models​

Use Descriptive Names:

  • Include domain: "Medical", "Legal", "Technical"
  • Include language: "EN", "NL", "FR"
  • Include specifics: "Cardiology", "Depositions", "IT Meetings"
  • Example: "Medical Cardiology EN"

Model Organization​

Create Separate Models For:

  • Different languages
  • Different domains (medical vs legal vs technical)
  • Different use cases

Training Data​

For Best Results:

  • Use high-quality audio recordings
  • Ensure audio matches your target use case
  • Include variety of speakers (if applicable)
  • More training data generally = better accuracy

Troubleshooting​

Model Won't Create​

Problem: Create button doesn't work

Solutions:

  • Ensure model name is filled in
  • Verify base language is selected
  • Check you have permission to create models
  • Try refreshing the page

Can't See Training Status​

Problem: Training status shows "Not Available"

Solutions:

  • Model may not have training initiated
  • Refresh the page
  • Check if model was properly created

Training Failed​

Problem: Status shows "Failed" (red)

Solutions:

  • Open model details to view error messages
  • Check training data quality and format
  • Verify sufficient training data provided
  • Contact support if issue persists

Model Deleted by Accident​

Problem: Accidentally deleted a model

Solutions:

  • Deletion is permanent, cannot be recovered
  • Will need to recreate and retrain the model
  • Consider implementing backup procedures for critical models

Next Steps​

Note: Custom model features and availability may vary based on your subscription plan. Training capacity, model limits, and advanced features depend on your organization's plan. Contact your administrator for details about custom model access.