A Survey On Deep Learning in Real Time Speech Packet-loss Concealment Methods

Lau, Michael

Permalink

https://hdl.handle.net/2142/110334 Copy

Description

Title

A Survey On Deep Learning in Real Time Speech Packet-loss Concealment Methods

Author(s)

Lau, Michael

Contributor(s)

Patel, Sanjay

Issue Date

2021-05

Keyword(s)

• Quality of Experience
• Machine Learning
• Packet Loss Concealment
• Audio Processing

Abstract

Packetloss occurring in Voice-Over-IP (VoIP) is the main source of degradation in quality of experience (QoE) during a call, especially in the world when we rely heavily on video conferencing where audio is arguable more important in terms of providing a high-quality experience. This thesis surveys the different methods proposed for this problem with respect to the real-time setting using deep learning and to compare the different metrics used and their performance. We found that recurrent models remain the most popular due to their ability to model long-term relationships, while auto-encoder and GANs are also used in packet-loss concealment (PLC). These findings suggest different possible methods that could potentially help solve the problem and improve the quality of experience overall.

Type of Resource

text

Language

en

Permalink

http://hdl.handle.net/2142/110334

Owning Collections