A Practical Introduction to Automatic Audio Segmentation Using Deep Learning

In Proceedings of: Late Summer School: Machine Learning for Language Analysis - September, 2018

Goal

Train a deep neural network on VGGish audio features to automatically segment an audio file with speech into speech and non-speech parts.

Setup

Please follow the instructions below in order.

Install and setup the following:

1. git
2. miniconda or anaconda (Python 3.x version)

Clone this repository WITH SUBMODULES (in the Terminal on Linux/macOS, or Git Bash on Windows):

# go to an existing directory where you'd like to clone this repository
cd ~

# clone this repo WITH SUBMODULES!
git clone --recurse-submodules https://github.com/motjuste/UoC-ml-school-2018

# change directory to the clone
cd UoC-ml-school-2018

Install the required packages using conda (Terminal on Linux/macOS, or Anaconda Prompt on Windows):

# change directory to the clone, if not already
# replace ~ below with where you clones this repository
cd ~/UoC-ml-school-2018

# update conda, just in case
conda update conda

# create a new conda environment with the required packages. May take time.
conda env create --file environment.yml

# activate the environment (also instructed at the end of the last step)
conda activate ml-school

Start jupyter notebook (Terminal on Linux/macOS, or Anaconda Prompt on Windows):

# change directory to the clone, if not already
# replace ~ below with where you clones this repository
cd ~/UoC-ml-school-2018

# activate the environment
conda activate ml-school

# start jupyter notebook
jupyter notebook

# if not opened automatically, copy the localhost URL and open it in a browser

Check setup by opening and following the instructions 00-check-setup.ipynb.