Web Interface Guide¶

Complete walkthrough of the WakeBuilder web interface.

Overview¶

WakeBuilder's web interface provides a user-friendly way to:

• View and manage wake word models
• Train new custom models
• Test models with live audio
• Monitor training progress

Navigation¶

The navigation bar at the top provides access to all sections:

Models Page (Home)¶

Model Dashboard¶

The home page displays all your wake word models as cards:

• Model name: The wake word
• Type badge: "Custom" or "Default"
• Metrics: Accuracy and F1 score
• Actions: Test, Download, Delete

Negative Data Panel¶

Shows the status of negative training data:

Click "Download Dataset" to get the UNAC corpus (~1.4GB).

Cache Panel¶

Shows negative cache status:

Click "Build Cache" to preprocess negative audio (~5 min).

Training Wizard¶

Step 1: Wake Word¶

Enter your desired wake word:

1. Wake word input: Type your trigger phrase
2. Model architecture: Select AST (default)
3. Advanced options: Click to expand for configuration

Advanced Options¶

Collapsible sections for customization:

Data Generation:

• Target positive samples
• Negative ratio
• Hard negative ratio
• TTS, real negatives, hard negatives toggles

Training Parameters:

• Batch size
• Max epochs
• Early stopping patience
• Learning rate
• Dropout
• Label smoothing
• Mixup alpha

Model Enhancements:

• Focal loss settings
• Self-attention
• SE block
• TCN block
• Classifier layers

Step 2: Recording¶

Record your voice saying the wake word:

1. Click and hold the record button
2. Speak your wake word
3. Release to stop
4. Review the recording (play, delete)
5. Repeat for 1-5 recordings

The waveform visualizer shows audio levels in real-time.

Step 3: Training¶

Monitor training progress:

Progress Section:

• Progress bar with percentage
• Current phase description
• Elapsed time

Training Data Panel:

• Recordings count
• Positive samples
• Negative samples
• Train/validation split

Hyperparameters Panel:

• All configured settings displayed

Metrics Panel:

• Current epoch
• Train loss, Val loss
• Val accuracy, Val F1

Charts:

• Loss history graph
• Accuracy/F1 history graph

Step 4: Results¶

View training results:

Performance Summary:

• Accuracy
• F1 Score
• Threshold
• Parameters count

Training Data Summary:

• Recordings provided
• Samples generated
• Epochs trained

Configuration Summary:

• All hyperparameters used

Threshold Analysis Chart:

• FAR/FRR curves
• Optimal threshold line

Actions:

• Test model
• Download model
• Train another

Test Page¶

Model Selection¶

Select a model to test from the dropdown menu.

Real-time Testing¶

1. Grant microphone access when prompted
2. Click "Start Listening"
3. Speak your wake word
4. Watch the detection indicator

Controls:

• Start/Stop button
• Threshold slider
• Confidence display

Detection Feedback¶

Keyboard Shortcuts¶

Mobile Support¶

The interface is responsive and works on mobile devices:

• Touch-friendly buttons
• Responsive layout
• Mobile recording support (where browser supports)

API Documentation¶

Click "API Docs" in the navigation to access:

Swagger UI¶

Interactive API documentation at /docs:

• Try endpoints directly
• View request/response schemas
• Authentication information

ReDoc¶

Alternative documentation at /redoc:

• Clean, readable format
• Grouped by tags
• Search functionality

Troubleshooting Interface Issues¶