Bounds on Agreement between Subjective and Objective Measurements

Objective estimators of multimedia quality are often judged by comparing estimates with subjective "truth data," most often via Pearson correlation coefficient (PCC) or mean-squared error (MSE). But subjective test results contain noise, so striving for a PCC of 1.0 or an MSE of 0.0 is neither realistic nor repeatable. Numerous efforts have been made to acknowledge and appropriately accommodate subjective test noise in objective-subjective comparisons, typically resulting in new analysis frameworks and figures-of-merit. We take a different approach. By making only basic assumptions, we derive bounds on PCC and MSE that can be expected for a subjective test. Consistent with intuition, these bounds are functions of subjective vote variance. When a subjective test includes vote variance information, the calculation of the bounds is easy, and in this case we say the resulting bounds are "fully data-driven." We provide two options for calculating bounds in cases where vote variance information is not available. One option is to use vote variance information from other subjective tests that do provide such information, and the second option is to use a model for subjective votes. Thus we introduce a binomial-based model for subjective votes (BinoVotes) that naturally leads to a mean opinion score (MOS) model, named BinoMOS, with multiple unique desirable properties. BinoMOS reproduces the discrete nature of MOS values and its dependence on the number of votes per file. This modeling provides vote variance information required by the PCC and MSE bounds and we compare this modeling with data from 18 subjective tests. The modeling yields PCC and MSE bounds that agree very well with those found from the data directly. These results allow one to set expectations for the PCC and MSE that might be achieved for any subjective test, even those where vote variance information is not available.

Unseen but not Unknown: Using Dataset Concealment to Robustly Evaluate Speech Quality Estimation Models

We introduce Dataset Concealment (DSC), a rigorous new procedure for evaluating and interpreting objective speech quality estimation models. DSC quantifies and decomposes the performance gap between research results and real-world application requirements, while offering context and additional insights into model behavior and dataset characteristics. We also show the benefits of addressing the corpus effect by using the dataset Aligner from AlignNet when training models with multiple datasets. We demonstrate DSC and the improvements from the Aligner using nine training datasets and nine unseen datasets with three well-studied models: MOSNet, NISQA, and a Wav2Vec2.0-based model. DSC provides interpretable views of the generalization capabilities and limitations of models, while allowing all available data to be used at training. An additional result is that adding the 1000 parameter dataset Aligner to the 94 million parameter Wav2Vec model during training does significantly improve the resulting model's ability to estimate speech quality for unseen data.

AlignNet: Learning dataset score alignment functions to enable better training of speech quality estimators

We develop two complementary advances for training no-reference (NR) speech quality estimators with independent datasets. Multi-dataset finetuning (MDF) pretrains an NR estimator on a single dataset and then finetunes it on multiple datasets at once, including the dataset used for pretraining. AlignNet uses an AudioNet to generate intermediate score estimates before using the Aligner to map intermediate estimates to the appropriate score range. AlignNet is agnostic to the choice of AudioNet so any successful NR speech quality estimator can benefit from its Aligner. The methods can be used in tandem, and we use two studies to show that they improve on current solutions: one study uses nine smaller datasets and the other uses four larger datasets. AlignNet with MDF improves on other solutions because it efficiently and effectively removes misalignments that impair the learning process, and thus enables successful training with larger amounts of more diverse data.