Urban Sound Classification

Python notebook for blog post Urban sound classification using Neural Network.

Python notebook for blog post Urban sound classification using Convolutional Neural Network.

Urban sound classification using RNN.

Project Goal:

This project uses laptop microphones and a vibrating Myo wristband to alert its wearer when a siren is detected. This is useful for deaf or hearing impaired drivers and pedestrians. It uses a small neural network based on the above blog posts to classify sound recorded from a microphone, sending a bluetooth signal to the myo wristband when a siren is detected. It currently works with 91% accuracy.

Setup and Tools Required

• Python 3.5
• Tensorflow
• Numpy
• Librosa
• myo sdk
• myo-python bindings for myo sdk
• pyaudio

Note that if training locally, the program will expect to find the data under data/audio and data/audio_balanced

Dataset

The UrbanSound8k dataset used for model training, can be downloaded from the following [link].

Another related dataset that can be used is Google's AudioSet.

Future steps:

• Standalone Android app
• Use a wristband with a microphone instead of myo
• Improve the neural network performance
• Improve the audio preprocessing
• Get more data, and try improving data (especially for training with vocal interference)