Relating Objective and Subjective Performance Measures for AAM-Based Visual Speech Synthesis

Abstract

We compare two approaches for synthesizing visual speech using Active Appearance Models (AAMs): one that utilizes acoustic features as input, and one that utilizes a phonetic transcription as input. Both synthesizers are trained using the same data and the performance is measured using both objective and subjective testing. We investigate the impact of likely sources of error in the synthesized visual speech by introducing typical errors into real visual speech sequences and subjectively measuring the perceived degradation. When only a small region (e.g. a single syllable) of ground-truth visual speech is incorrect we find that the subjective score for the entire sequence is subjectively lower than sequences generated by our synthesizers. This observation motivates further consideration of an often ignored issue, which is to what extent are subjective measures correlated with objective measures of performance? Significantly, we find that the most commonly used objective measures of performance are not necessarily the best indicator of viewer perception of quality. We empirically evaluate alternatives and show that the cost of a dynamic time warp of synthesized visual speech parameters to the respective ground-truth parameters is a better indicator of subjective quality

Expressive Modulation of Neutral Visual Speech

“…comparing feature trajectories between synthesised and ground truth data, or calculating RMS error between synthesised and ground truth meshes etc. However, Theobald and Matthews [2012] show that objective measures are not necessarily a good indicator of the subjective perception of naturalness in a synthesis technique. RMS error seems a particularly poor choice since it averages across an entire sequence, whereas the authors show that an artefact in a single frame of a sequence can lead to the entire sequence being perceived as bad.…”

Visual Speech Synthesis Using Dynamic Visemes, Contextual Features and DNNs

“…Samplebased approaches concatenate visual speech units contained in a database, where the units might be fixed-length (e.g. phonemes, visemes, or words [4,5,6,7]) or of variable length [8,9,10]. A cost function, based on phonetic context and smoothness of concatenation, is then minimised to find the set of units which form the animation.…”