Investigating Range-Equalizing Bias in Mean Opinion Score Ratings of Synthesized Speech

TL;DR

This paper investigates how range-equalizing bias influences MOS ratings in synthesized speech evaluations by conducting systematic listening tests to understand and quantify this bias.

Contribution

It provides a systematic analysis of range-equalizing bias in MOS tests for synthesized speech, highlighting its impact on evaluation results.

Findings

Range-equalizing bias causes listeners to use the full scoring range regardless of sample quality.

Zooming in on higher-quality samples reveals the extent of bias in MOS ratings.

Quantitative analysis of bias effects informs better evaluation practices.

Abstract

Mean Opinion Score (MOS) is a popular measure for evaluating synthesized speech. However, the scores obtained in MOS tests are heavily dependent upon many contextual factors. One such factor is the overall range of quality of the samples presented in the test -- listeners tend to try to use the entire range of scoring options available to them regardless of this, a phenomenon which is known as range-equalizing bias. In this paper, we systematically investigate the effects of range-equalizing bias on MOS tests for synthesized speech by conducting a series of listening tests in which we progressively "zoom in" on a smaller number of systems in the higher-quality range. This allows us to better understand and quantify the effects of range-equalizing bias in MOS tests.