spoken-digits-recognition

University of Toronto School of Continuing Studies

Term project for Machine Learning course

Group: Ankur Tyagi, Haitham Alamri, Rodolfo de Andrade Vasconcelos

Professor: Matthew MacDonald

Presentation

Business Problem

• Inference of the digits as said by users on phone
• Biometric authentication using speech dataset

Prerequisites

1) if running locally, have Python 3.7 and Junyper installed

How to run

Local Jupyter Notebook

1. clone this repository
2. ensure you have the correct version of the Python libraries stated in the file src/lib_version Example:

   python -m pip install -U <library>==<version>

3. Run junyper nootebook

   junyper notebook

4. Open src/spoken-digits-recognition.ipynb or src/speaker-recognition.ipynb
5. Run all Cells

Colab Jupyter Notebook

1. Open
2. Upload the CSV files
3. Run all Cells

Contents

• README.md: This file, explaining the project

• UofT_Final_project.pdf: Project presentation in PDF format

• UofT_Final_project.pptx: Project presentation in Power Point format

• spoken_digits_comparison.pdf: Recordings comparisson for each digit for the six speakers

• data/*: wav files with English spoken digits from 0 to 9

• src/lib_version: Python libraries version used in this project

• src/more_test.csv: Features of the files in data/recordings/moreSpeakersTest

• src/more_train.csv: Features of the files in data/recordings/moreSpeakersTrain

• src/speaker-recognition.ipynb: Junyper Notebook with Keras Neural Network model able to recognize the speaker of English digits (Jackson, Nicolas, Theo, Ankur, Caroline and Rodolfo)

• src/speaker-recognition.pdf: PDF version of an execution of speaker-recognition.ipynb

• src/spoken-digits-recognition.ipynb: Junyper Notebook with Neural Network model able to recognize English spoken digits

• src/spoken-digits-recognition.pdf: PDF version of an execution of spoken-digits-recognition

• src/test.csv: Features of the files in data/recordings/test

• src/train.csv: Features of the files in data/recordings/train

Recording Data

The files stored in data/recordings/test and data/recordings/train were downloaded from FSDD [Ref 1 and 2]:

Ankur, Caroline and Rodolfo provided the recordings stored in data/recordings/moreSpeakersTest and data/recordings/moreSpeakersTrain

References

1. https://github.com/Jakobovski/free-spoken-digit-dataset
2. https://github.com/moebg/spoken-digit-recognition