Syntax Speaker Prediction

This repository contains notebooks used to train a model to predict the speaker (Scott or Wes) in one second cilps of their podcast, syntax.fm. There are accompanying files for using Prodigy to label the data required to build the model. There is a brief description of the purpose of each notebook as the first cell.

The Results

Total Podcast Time: 3 days, 15 hours, 56 minutes, 39 seconds

Wes: 2 days, 1 hours, 51 minutes, 11 seconds

Scott: 1 days, 11 hours, 51 minutes, 45 seconds

Non-speaking time (crosstalk, laughing, intros): 0 days, 2 hours, 13 minutes, 43 seconds

Cumulative Speaking Time

Each vertical line is the start of a new episode.

Helpful tools:

• Prodigy for data labeling
• Pydub for slicing audio
• Librosa for audio analysis
• pandas for data manipulation
• scikit-learn for machine learning
• chartify for plots