EMERGENCY_VEHICLE_DETECTION_SYSTEM
ABOUT
This repository contains implementation of real world problem of Emergency Vehicle Siren Detection .Underlying idea is to process the audio signal of siren to extract its various features and train the model using best suited Machine Learning algorithm in order to achieve end result of maximum accuracy. Numerous other techniques were used before applying KNN ,ANN for a conclusive evidence to attain good accuracy.
IMPORTANT LIBRARIES
1.pandas
2.Numpy
3.Tensorflow
4.matplotlib
5.librosa
6.sci-kit learn
ROADMAP
1st approach
1.From audio files we need to extract important features i.e. MFCCs, Spectral Centroid, Zero Crossing Rate, Chroma Frequencies, Spectral Roll-off.
2.Once the features have been extracted, they can be appended into a CSV file.
3.Various classification algorithms can be used to get better accuracy .
**some important feature are:
2nd approach 1.We need to convert the audio files into PNG format images(spectrograms).
2.Train a CNN model for classification using these spectrogram images.
some smaple pictures of spectrogram
DATA
features_information
Data preparation
various algorithms
It was observed that in general emergency signals having lot of spurious noise mixed along with them were consistently misclassified by the ANN. Also, in many cases the ANN failed to learn the features of emergency signals that were unique which led to misclassification. A possible solution to this issue would be to use more training data with samples that are unique. Another solution would be to use a wider variety of features to get that could be helpful in distinguishing between emergency and non-emergency signals.