Project of Module EG2303:
Automatic Speech Recognition System
October 2024
0.1 Description
The goal of this group project is to provide students with hands-on expe- rience in the principles and concepts underlying signals and systems. The students are expected to develop an automatic speech recognition system designed to recognise spoken numbers. The emphasis in this project will be on the manual implementation of noise filtering and feature extraction in the frequency domain, without relying on external libraries. The speaker will be always the one in Google Translator (www.translate.google.com), and the digits of each number are always pronounced separately (this can be obtained by using spaces between digits). The system should receive a voice pronouncing a 10 digit number and show the whole number in the text format. As an example, when the input is “two three nine zero” in the voice format, the output should be “2 3 9 0” in the text format. The system should be able to work correctly even when the input voice is noisy. The developed systems will be tested during the presentation session in a lecture room in real time.
Note that the development of such an automatic speech recognition sys- tem can normally be divided into several key steps: (i) data collection (recording a 10 digit number pronounced by Google Translator and sav- ing the recording in an audio format), (ii) preprocessing the data (trimming the audio file and splitting it into 10 diferent signals), (iii) feature extraction (implementing a noise cancelling filter and obtaining a feature variable in the frequency domain for each signal), and (iv) classification (comparison of the feature variable of each signal with those in a voice bank, and identifying the correct digit).
0.2 Groups
There will be 9 groups each consists of 6 members. Groups are allocated randomly, and the list of the members of each group will be available on Moodle. The members of each group are expected to participate equally in the project. Each group must work independently, and any exchange of information between diferent groups is forbidden.
0.3 Assessment
Each group is supposed to prepare a report and present its developed system so that its correctness can be assessed. The report should be maximum 5 pages and the presentation time will be limited to maximum 10 minutes. Note that this project accounts for 50% of the final mark of each student. The performance of each group will be assessed based on the following cri- teria:
• Clarity and Coherence (10%)
• Equal Participation (10%)
• Comprehensiveness (40%)
• Correctness (40%)
Note that, each member of a group will be able to influence the marks of other members in the same group through a peer review process, described in the attached document.
0.4 Deadline
The deadline for submission of reports will be 9th of December, 2024 (i.e., the date of the last lecture of the module). Each group must make sure to submit its report on Moodle by the deadline. The presentations will be held on the same date during the lecture session.