Vector quantization (VQ)–based generative DNN models for low delay speech and audio coding.

This is a Master's Thesis for the last semester of the Master's Programme In Computer Engineering (AI, Vision, and Sound) at Aalborg University.

Author

• Lukas Bisgaard Kristensen

Supervisors

• Jan Østergaard
• Mohammad Bokaei

Installing

Installing the packages required: pip install -r requirements.txt

Dataset

The dataset used for running the models is the LibriTTS corpus (cleaned), which can be accessed through the link https://openslr.org/60/.

Structure

The models folder contains three different models (AE100, AE25, and AE-VQ) and a main.py, which gathers information about the dataset.

Results

For each model and packet loss variation, a sample was plotted together with a sample that can be played compared to the ground truth and can be seen in the Results folder. The baseline performance for the models without packet loss can be seen in the image below: