Perceptual Evaluation of Audio Quality

Description

The Perceptual Evaluation of Audio Quality (PEAQ) is a standardized algorithm defined by ITU-R BS.1387 and adopted by 3GPP for objectively measuring the quality of processed audio signals. It operates by comparing a processed (degraded) audio signal to an original reference signal. The core of PEAQ is a psychoacoustic model that mimics the human auditory system, analyzing signals in terms of loudness, masking effects, and critical bands. The algorithm extracts specific perceptual features, known as Model Output Variables (MOVs), from both signals. These MOVs capture differences in aspects like noise loudness, modulation, and harmonic structure. A neural network or a simpler cognitive model then maps these extracted perceptual differences into a single objective quality score, typically the Objective Difference Grade (ODG), which ranges from -4 (very annoying impairment) to 0 (imperceptible impairment). This allows for a reliable prediction of a subjective Mean Opinion Score (MOS). In 3GPP, PEAQ is applied within test specifications to evaluate the performance of audio and speech codecs (e.g., EVS, AMR-WB) under various network conditions like packet loss and jitter. The process is fully automated, enabling reproducible and efficient quality assessments during codec development, selection, and network planning. Its integration into 3GPP specifications ensures that audio quality metrics are consistent and comparable across different implementations and vendors, forming a critical part of the end-to-end quality assurance framework for multimedia services.

Purpose & Motivation

PEAQ was created to address the need for an efficient, reliable, and standardized method to assess audio quality objectively, replacing or supplementing costly and time-consuming subjective listening tests (like ACR tests). Subjective tests require panels of human listeners, are not easily repeatable, and are impractical for automated testing or regression analysis during codec development. Prior to PEAQ, objective metrics like Signal-to-Noise Ratio (SNR) were often used, but these are poor predictors of perceived quality because they do not account for the non-linear and frequency-dependent characteristics of human hearing. The motivation was to develop a tool that could accurately predict listener opinions based on a computational model of psychoacoustics. Its adoption into 3GPP standards, starting with Rel-8, provided a common benchmark for evaluating the performance of evolving voice and audio codecs across generations of mobile networks (3G, 4G, 5G). This ensures that quality is maintained and improved as networks introduce new compression techniques and handle diverse audio content, from narrowband speech to high-definition music.

Key Features

• Objective quality prediction based on psychoacoustic models
• Outputs an Objective Difference Grade (ODG) correlating to subjective MOS
• Compares processed audio to an original reference signal
• Extracts Model Output Variables (MOVs) for perceptual differences
• Enables automated, reproducible testing of codec performance
• Standardized methodology (ITU-R BS.1387) adopted by 3GPP

Evolution Across Releases

Defining Specifications