Predicting pairwise preferences between TTS audio stimuli using parallel ratings data and anti-symmetric twin neural networks
This work addresses the challenge of automating subjective listening tests for TTS systems, which is incremental as it builds on existing methods by focusing on pairwise comparisons rather than individual ratings.
The paper tackled the problem of predicting subjective preferences between two text-to-speech audio stimuli by using anti-symmetric twin neural networks trained on pairwise preference data derived from MUSHRA ratings, achieving results that compare favorably to a state-of-the-art mean opinion score prediction model.
Automatically predicting the outcome of subjective listening tests is a challenging task. Ratings may vary from person to person even if preferences are consistent across listeners. While previous work has focused on predicting listeners' ratings (mean opinion scores) of individual stimuli, we focus on the simpler task of predicting subjective preference given two speech stimuli for the same text. We propose a model based on anti-symmetric twin neural networks, trained on pairs of waveforms and their corresponding preference scores. We explore both attention and recurrent neural nets to account for the fact that stimuli in a pair are not time aligned. To obtain a large training set we convert listeners' ratings from MUSHRA tests to values that reflect how often one stimulus in the pair was rated higher than the other. Specifically, we evaluate performance on data obtained from twelve MUSHRA evaluations conducted over five years, containing different TTS systems, built from data of different speakers. Our results compare favourably to a state-of-the-art model trained to predict MOS scores.