Name		Name	Last commit message	Last commit date
parent directory ..
README.md		README.md
Visualization_sushi.ipynb		Visualization_sushi.ipynb
accuracy_analysis_my_home.ipynb		accuracy_analysis_my_home.ipynb
accuracy_analysis_my_life_environment.ipynb		accuracy_analysis_my_life_environment.ipynb
input.png		input.png
key_word_detection_cnn_success_1.ipynb		key_word_detection_cnn_success_1.ipynb
key_word_detection_dnn_success_1.ipynb		key_word_detection_dnn_success_1.ipynb
layer0.png		layer0.png
layer1.png		layer1.png
layer2.png		layer2.png
layer3.png		layer3.png
layer4.png		layer4.png
layer5.png		layer5.png
musical_instruments_cnn_success_1.ipynb		musical_instruments_cnn_success_1.ipynb
my_home_cnn_success_1.ipynb		my_home_cnn_success_1.ipynb
my_life_environment_cnn_success_1.ipynb		my_life_environment_cnn_success_1.ipynb
run_inference_music.bat		run_inference_music.bat
run_inference_restaurant.bat		run_inference_restaurant.bat

README.md

Deep learning on Keras/TensorFlow

The results: about 90% accuracy has been achieved, so it is satisfying.

The trained CNN model will work OK if the following condition is satisfied:

use the same MEMS mic with the same filter setting (incl. the same frequency response) on the edge device (STM32L4)
same environment
less surrounding noise
same window setting