https://github.com/radlfabs/svd-spoken-digits-classifier

SVD Classifier: Implementation from scratch. Trained and evaluated with AudioMNIST.
https://github.com/radlfabs/svd-spoken-digits-classifier

Last synced: 26 days ago
JSON representation

SVD Classifier: Implementation from scratch. Trained and evaluated with AudioMNIST.

• Host: GitHub
• URL: https://github.com/radlfabs/svd-spoken-digits-classifier
• Owner: radlfabs
• License: mit
• Created: 2022-09-26T08:57:09.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2023-10-30T18:06:54.000Z (about 1 year ago)
• Last Synced: 2024-10-28T14:18:50.361Z (2 months ago)
• Language: Python
• Homepage:
• Size: 1.79 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# SVD based Classifier for SpeechRecognition on AudioMNIST

This repo hosts our implementation of the classical SVD-based MNIST digit classifier but applied to AudioMNIST. This project has been conducted during the final project in the course Applied Mathematics at University of Applied Science Cologne. 

# Getting Started

For testing purposes we provide a small subset of AudioMNIST. It has been randomly created with `dev_dataset.py` and 

represents a minimal working example with 30 audio files in each digit class(300 total). 

It is important to point out that the classification error on this data set is not representative. 

Nonetheless, it is a useful way to test the main program without downloading the full data set. 

In addition, the main program will only classify 5 test samples by default to limit computation time. 

So testing will most likely stop before converging to the lowest possible error rate. 

This program was created with Python 3.10.4.

1. Clone this repo and `cd` into the directory.

2. Create a new python venv and activate it.

3. Install required packages with `pip install -r requirements.txt`. 

4. If you would like to perform a complete classification run, clone this [repo](https://github.com/soerenab/AudioMNIST.git) and move the 'data' folder to your local repo.

5. Consecutively, you can run the main program (`number_classifier.py`) for testing or grading purposes from the 

   command line. `cd` to the unzipped directory and run the command `py -m number_classifier`. 

   Of course, you can run the file in any Python IDE of your liking.

6. Additional plots can be generated by running the following prompt: `py -m util`

   They will be saved as pdf-files to a plots-subdirectory.