Background
The goal of this homework is to get some experience building a dataset from online sources, cleaning it and performing some basic visualizations. To do so, you will make use of scraping libraries, regex and matplotlib .
Specifically, we will be working with an audio dataset provided by Google Research described here. However, we will assume it is provided in somewhat of a raw format: a CSV file with YouTube IDs, timestamps and tags.
It will be your job to take this CSV, download the associated audio, format them to only include the relevant segment and clean up the data so it can be used for some downstream ML task (for example, training an audio classifier or a generative model to create similar audio samples).
In addition, you will be visualizing various aspects of the dataset to gain a better understanding of the dataset.
Getting started
This assignment has 2 components that need to be completed: - The .py files contain functions that should be filled in as specified in the comments - The visualization.ipynb contains cells that need to be filled in and run as well as some basic questions that should be answered in markdown cells.
Start by setting up a virtual environment as you did in the previous homework and install the contents of requirements.txt .
It is then recommended to complete the questions in order (starting from the .py file then the corresponding sections of the notebook) as the output of previous questions is sometimes used for later questions.
Questions
1. Understanding and visualizing the dataset
To begin, we want to make audio_segments.csv more human readable and better understand the distribution of labels.
Complete the functions in q1.py then fill in and run the cells in visualization.ipynb under the section Question 1.
Looking at audio_segments.csv , the labels for each video correspond to the ID of the label and not the actual label name.
2. Downloading and processing data
Now that we've cleaned up the .csv , it's time to get at what we're really interested in: the audio. Complete the functions in
q2.py which will be the building blocks of our small data processing pipeline. One function should download the audio and the other should cut it to only include the segment mentioned in the .csv
3. Building the dataset using a pipeline
With these building blocks, we will build a very small data pipeline to download and process the entire dataset Complete the functions in q3.py then run the cell in visualize.ipynb
4. Visualizing audio
Now that the segments are downloaded, complete the cells in visualize.ipynb to listen to and visualize some of the audio samples.
References
https://research.google.com/audioset//download.html
https://regexone.com/
