Mean Opinion Score – Listening Quality Objective

Description

Mean Opinion Score – Listening Quality Objective (MOS-LQO) is a computed metric that represents an estimated subjective listening quality score. Unlike the classic MOS derived from human panels, MOS-LQO is generated entirely by software algorithms defined in ITU-T standards, such as P.862 (PESQ) and P.863 (POLQA). These algorithms are 'full-reference' models, meaning they compare a degraded speech signal (as received after passing through the network and codecs) with the original, pristine reference signal. By analyzing specific perceptual differences between the two signals—such as distortions, noise, delays, and codec artifacts—the algorithm calculates a score that highly correlates with average human subjective ratings (MOS). The output is a number typically on a scale similar to MOS (e.g., 1.0 to 4.5 or 1.0 to 5.0, depending on the algorithm).

The architecture of an MOS-LQO measurement system involves a test setup where a known reference speech file is injected into the system under test (e.g., a voice call path over an IMS network). The output signal is captured and time-aligned with the reference. The objective model (e.g., POLQA) then performs a sophisticated perceptual analysis. It models the human auditory system, applying concepts like loudness perception, masking effects, and cognitive modeling to identify impairments that a listener would notice. The model aggregates these perceptual disturbances into a single numerical score, the MOS-LQO. This process is fully automated, repeatable, and can be executed in real-time or on recorded files, making it vastly more scalable than subjective testing.

Within 3GPP, MOS-LQO is specified as the primary method for conformance testing and performance benchmarking of speech codecs, voice over IP (VoIP) systems, and entire network paths. Specifications define test conditions (e.g., background noise levels, packet loss profiles) and minimum required MOS-LQO values that a codec or device must achieve. For instance, 3GPP TS 26.179 defines Enhanced Voice Services (EVS) codec performance requirements using MOS-LQO targets under various error conditions. Network operators also use MOS-LQO measurements from drive tests or passive probes to monitor the end-user voice quality experience across their network, enabling them to identify quality degradation and optimize network parameters proactively.

Purpose & Motivation

MOS-LQO was developed to overcome the major practical limitations of subjective MOS testing: it is extremely time-consuming, expensive, and not suitable for live network monitoring or rapid product development cycles. While subjective MOS provides the definitive quality benchmark, it cannot be used for continuous quality assurance in operational networks or for testing every possible network condition. The purpose of MOS-LQO is to provide a practical, automated, and accurate proxy for subjective listening quality.

Its creation was motivated by the need for standardized, instrument-based quality assessment that could be embedded in test equipment, network probes, and device certification labs. By providing an objective prediction that strongly correlates with human perception, MOS-LQO enables engineers to make quality comparisons and assessments at scale and in real-time. It addresses the critical industry requirement for quantifying Quality of Experience (QoE) in a way that is both scientifically valid and operationally feasible, forming the basis for modern voice service quality management and benchmarking in 2G, 3G, 4G, 5G, and OTT communication services.

Key Features

• Objective prediction of subjective listening quality (MOS) using algorithmic models
• Based on full-reference comparison between original and degraded speech signals
• Implements perceptual models of the human auditory system (e.g., in POLQA)
• Enables automated, repeatable, and scalable speech quality testing
• Used for codec conformance testing, device certification, and network quality monitoring in 3GPP
• Output is a numerical score highly correlated with results from subjective listener panels

Evolution Across Releases

Defining Specifications