Inspired by the people at the grocery store thumping on watermelon. You're looking for a hollow sound to indicate that the watermelon is ripe. In this project I went to HEB and collected about 70 samples thumping on watermelons and used it to train a neural network through sklearn. I first used librosa library to create a spectrogram and extracted features of each sample (a "thump" in this case) such as Mel-frequency cepstral coefficients and chroma. The data is transformed into a np.array and put through a multi-layer perceptron with one hidden layer.
After the neural net is trained, the user can input a short audio file if their watermelon thump and the program will tell you if the thump is the hollow sound you want for a ripe melon. Especially useful for the hearing impaired, or just people who can't quite distinguish a ripe sound from a non-ripe one.