TinyML: Wake Word Detection

Team Member:

Team Member:

Han Jiang

Goal:

Goal:

After running the example of detect "Yes" and "No", I trained a model that defined "On" and "Off" as the keywords and tried to classify these words. When individuals utter words near the Arduino board, it will automatically detect them.

Process:

Process:

Training model in Colab.

• Training model in Colab.

Configure training: os.environ["WANTED_WORDS"] = "on, off".

• Configure training: os.environ["WANTED_WORDS"] = "on, off".

Traing the model and monitor its process by Tensorboard

• Traing the model and monitor its process by Tensorboard

Freeze the model.

• Freeze the model.

Convert to Tensorflow Lite and then create a C array.

• Convert to Tensorflow Lite and then create a C array.

Replacing the model: replace the C array in the micro_features_model.cpp

• Replacing the model: replace the C array in the micro_features_model.cpp

Updating Labels: update the labels in the micro_features_micro_model_settings.cpp

• Updating Labels: update the labels in the micro_features_micro_model_settings.cpp

Updating command_responder.cc

• Updating command_responder.cc

Result:

Result: