9_Urban_Sound_Classification

My solution to solve a simple practice Audio classification challenge on Analytics Vidhya's website - Urban Sound classification. Here is where I stand with few hours of hacking -

Problem statement -

Audio classification in 10 different classes.

Data

This dataset contains 8732 labeled sound excerpts (<=4s) of urban sounds from 10 classes: air_conditioner, car_horn, children_playing, dog_bark, drilling, enginge_idling, gun_shot, jackhammer, siren, and street_music.

Approach

Using fastaiaudio to convert audio into Mel Spectorgram and then use a pretrained Resnet18 model to classify in 10 different categories. Refer to the UrbanSoundClassification.ipynb notebook.