Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: THEend8_
Neural Networks and Deep Learning
Deadline: 18th March 2024
● This assignment is to be done individually. You can discuss the questions with others but your submission must be your own unique work.
● Data files and other supporting code for both parts are found in the folder ‘Programming Assignment’ under ‘Assignments’ on NTULearn. You should use the helper codes to begin and provide answers to the assignment. Please follow the formats given in helper codes to fill in your solutions. Those who do not follow the specified format for the answers risk losing marks for presentation.
● The assessment will be based on the correctness of the codes and solutions. Each part carries 45 marks, and 10 marks are assigned to the presentation and clarity of the solutions. The total number of marks is 100.
● You do not need GPUs for this assignment. Local PCs will be sufficient. Attempt your assignments using Jupyter Lab / Notebook with version 3 and greater.
● TAs Mr. Feng Ruicheng ([email protected] ), Ms. Liang Zhenxin ([email protected] ), and Mr. Liao Chang ([email protected]tu.edu.sg ) are in charge of this assignment.
● For any inquiries, please either contact TAs listed above or post in the Discussion Board of NTULearn.
Submission procedure
● Complete both parts A and B of the assignment and submit your solutions online via NTULearn before the deadline.
● All submissions should be within the notebooks provided. Do not include data nor model checkpoints in your submission. Submit 8 notebooks in this format:
o For part A
. <lastname>_<firstname>_Part_A_1.ipynb . <lastname>_<firstname>_Part_A_2.ipynb . <lastname>_<firstname>_Part_A_3.ipynb . <lastname>_<firstname>_Part_A_4.ipynb . common_utils.py
o For Part B
. <lastname>_<firstname>_Part B 1.ipynb . <lastname>_<firstname>_Part B 2.ipynb . <lastname>_<firstname>_Part B 3.ipynb . <lastname>_<firstname>_Part B 4.ipynb
● Late submissions will be penalized: 5% for each day up to three days.
Part A: Classification Problem
Part A of this assignment aims at building neural networks to perform. polarity detection from voice recordings, based on data in the National Speech Corpus, which is obtained from https://www.imda.gov.sg/how-we-can-help/national-speech-corpus
The National Speech Corpus is an initiative by the Info-Communications and Media Development Authority, and it is the first large scale Singapore English corpus. Within the dataset, there are 6 parts. In the fifth segment, speakers are made to communicate in several different styles, including Positive Emotions and Negative Emotions. The original recordings are approximately 20 minutes long. Using the librosa library, the recordings are split into shorter segments and preprocessed to features such as chromagrams, Mel spectrograms, MFCCs and various other features.
The preprocessed CSV file is provided in this assignment. We will be using the CSV file named simplified.csv, which is both provided to you. The features from the dataset are engineered. The aim is to determine the speech polarity of the engineered feature dataset. The csv file is called simplified.csv with a row of 77 features that you can use, together with the filename. The “filename” column has the labels associated with them.