Custom Models
As a reseller, you can create custom speech recognition models and share them with your customer organizations. Custom models improve transcription accuracy for specific industries, accents, terminology, or use cases.
Overview
Custom models are specialized speech recognition models trained on your specific vocabulary and terminology, providing higher accuracy for domain-specific content. Create once and share across multiple customer organizations.
Creating Custom Models
When to Create Custom Models
Consider custom models when:
- Multiple customers share an industry or domain
- Standard models struggle with specific terminology
- Customers have consistent accent or speaking style
- High-value customers need exceptional accuracy
Good Use Cases: Medical, legal, financial services, technical support, specific accents or dialects.
Create Model
- Go to Reseller Dashboard
- Click Custom Models
- Click Create Custom Model
- Fill in details:
- Model Name - Descriptive name (e.g., "Medical EN-US")
- Description - What it's optimized for
- Base Language - Primary language
- Domain - Industry or use case
- Click Create Model
Upload Training Data
Requirements: Minimum 10 hours of clear audio (WAV, MP3, or FLAC) with accurate word-for-word transcripts. Each audio file must have a matching transcript file.
- In the custom model page, click Upload Training Data
- Choose upload method:
- Drag and drop - Upload multiple files
- Folder upload - Upload entire folder structure
- TUS upload - For large datasets (>1GB)
- System validates:
- Audio quality
- Transcript matching
- File formats
- Review validation report
- Fix any issues
- Click Confirm Upload
Add Vocabulary & Train
Add custom terms (with optional pronunciation and context), then start training. Training typically takes 24-72 hours.
Test the Model
After training, test with new audio files and compare accuracy to the standard model. Accept to deploy, or retrain with additional data if needed.
Managing Custom Models
Model Dashboard
View all your custom models:
Model Cards Show:
- Model name and description
- Training status (training, active, archived)
- Accuracy metrics
- Organizations using it
- Total usage (minutes transcribed)
- Last updated date
Model Details
Click on a model to see:
Performance Metrics
- Word Error Rate (WER)
- Accuracy vs. standard model
- Performance by speaker
- Common errors
Usage Statistics
- Total transcription minutes
- Organizations using model
- Users per organization
- Usage trends
Training Information
- Training data size
- Training duration
- Vocabulary size
- Last training date
Update Models
Add new training data or vocabulary, then retrain. Each training creates a new version that you can test before deploying to customers.
Sharing Models with Organizations
Assign Model to Organization
Give customers access to custom models:
- Go to organization details
- Click Custom Models tab
- Click Assign Model
- Select model from list
- Set as:
- Default - Used automatically for all transcriptions
- Optional - Customer chooses when to use
- Click Assign
Multiple Organizations
Assign same model to multiple customers:
- Go to custom model details
- Click Manage Organizations
- Select organizations to grant access
- Choose default or optional
- Click Apply to Selected
Use Cases:
- Industry-specific model for all medical customers
- Regional accent model for customers in same area
- Shared terminology model for related businesses
Usage Permissions
Control how organizations use models:
View Only
- See model exists
- View accuracy metrics
- Cannot modify or retrain
Use in Transcription
- Select model for transcriptions
- View performance
- Cannot modify model
Contribute Training Data
- Upload additional training data
- Help improve shared model
- Collaborative improvement
Remove Access
Revoke model access:
- Go to organization details → Custom Models
- Find model to remove
- Click Remove Access
- Choose:
- Immediate - Remove now
- After date - Schedule removal
- Confirm removal