Audio Streaming Quality Metric

Iyer, Kishore

Permalink

https://hdl.handle.net/2142/110316 Copy

Description

Title

Audio Streaming Quality Metric

Author(s)

Iyer, Kishore

Contributor(s)

Patel, Sanjay

Issue Date

2021-05

Keyword(s)

• streaming audio quality assessment
• MOS

Abstract

This research project was part of LiveSensus, a project within the Alchemy Technology Foundry with the goal of assessing the audio quality of livestreams and video conferencing. To accomplish this, multiple detectors were created to assess various types of streaming issues such as packet loss and background noise. The team also created a survey which included multiple audio clips to be scored based on the quality of the clip. This gave us more insight into the subjective evaluation of audio quality. The purpose of the research project was to aggregate the results from the individual detectors and output a final quality score from 1 to 5. This was done by training and testing a machine learning model on the dataset of audio clips used in the aforementioned survey. The outputs of the individual detectors were used as features, and the ratings by those who took the survey were used as labels. Problems to solve for this project included preprocessing and normalizing the individual detector outputs and survey ratings, selecting the optimal model, and tuning hyperparameters. Many different models were tested, and the best performing model currently has a mean absolute error of 0.3 on a scale from 1 to 5.

Type of Resource

text

Language

en

Permalink

http://hdl.handle.net/2142/110316

Owning Collections