Spoken Digit Classification

• Description
  Implement a classifier able to predict which digit is pronounced in a short audio excerpt.
• Input
  The dataset used is the Free Spoken Digit Dataset (FSDD). In the folder recodings you will find the audio files named in a specific format. Please read the READ ME distributed with the dataset. The results of the classification must be reported as a confusion matrix and, optionally, other metrics of your choice.
• Output
  • a brief presentation of your work (max 5 minutes) that will be given to the class
  • a more detailed report in which you illustrate and explain every step of your classification system and in which the results are shown and commented (max 8 pages) to be delivered by May 17th.
  • a link to a repository containing the code (e.g. on GitHub) with minimal comments.

Tasks:

• Preprocessing
• Feature selection
• Dataset split
• Feature extraction
• Feature selection
• Classification
• Performance evaluation

Features

Mel-frequency cepstrum coefficients
Linear Predictive Coding
Phoneme detection
HiddenMarkovModels
Image processing on spectrogram