Machine Learning model is necessary for baby crying detection. In this project, the task is binary classification: given 160 different features, we need to classify whether it’s 1 (baby crying sound) or 0 (other sound). To achieve this, several models can be used, like SVM, ANN and GMM. The accuracy of sound detection is related to many factors, for example, extracted features, audio quality, machine learning model and dataset used to train the model.
In this project, we use AWS machine learning to detect baby crying sound. The dataset for model training contains 140 pieces of baby crying sound and 275 pieces of other sound. Baby crying sound is similar but other sound contains street noise sound; people talking sound; sirens sound some other noise. The evolution of this model shows high accuracy (above 98%), because baby crying sound is relative unique comparing to other sound so it’s easy to detect.
Finally, with this model, we can make classification to a real time sound. After recording sound inside the car, a .wav file is made and its features will be extracted. By feeding these features to the model, we can get the final classification and know whether the sound is from baby crying.
In this project, we use AWS machine learning to detect baby crying sound. The dataset for model training contains 140 pieces of baby crying sound and 275 pieces of other sound. Baby crying sound is similar but other sound contains street noise sound; people talking sound; sirens sound some other noise. The evolution of this model shows high accuracy (above 98%), because baby crying sound is relative unique comparing to other sound so it’s easy to detect.
Finally, with this model, we can make classification to a real time sound. After recording sound inside the car, a .wav file is made and its features will be extracted. By feeding these features to the model, we can get the final classification and know whether the sound is from baby crying.