Audio Classification — Model Analysis

Input Files Required

Input	Description
Model (.h5)	Trained TensorFlow/Keras model
X Test (.npy)	Test feature dataset
Y Test (.npy)	Test label dataset

Model Visualisation

Load the .h5 model and click Visualize Model. The architecture diagram displays:

• Input shape of the model
• Layer-by-layer architecture based on the selected model
• Input and output shape of each layer
• Presence of QuantizeWrapperV2 (if Quantization-Aware Training (QAT) was used)
• Final number of output classes

Run Conversion

Converts the .h5 model to a deployable NDP firmware package (.synpkg).

1. Browse to the .h5 model file
2. Select X_test.npy and Y_test.npy
3. Click Run Conversion and monitor the console

During conversion the tool:

• Quantizes weights to INT8
• Converts layers to Syntiant hardware operations
• Maps model into NDP memory banks
• Generates .synpkg output files

Run Simulation

Validates that the converted NDP model produces the same predictions as the original Keras model.

1. Click Run Simulation and monitor logs

Run Comparison

Compares original Keras and NDP simulated model to verify conversion preserved accuracy.

1. Ensure both conversion and simulation are complete
2. Click Run Comparison

Results shown in tabs: Generated Files, Prediction Table, Confusion Matrix, Time Series Graph.