Intrinsic direction-dependent velocities of articulators
Final Report Abstract
The main goal of this project was to find out whether there are consistent intrinsic differences (that do not underlie conscious control) of the velocities of articulators for the movement in different directions (e.g., raising vs. lowering movements of the larynx, or vocal fold abduction vs. adduction). These potential differences have been explored in four separate speech production studies with novel experimental designs and dedicated measurement techniques (ultrasound imaging, electromagnetic articulography, laryngeal video endoscopy). The results of the experiments suggest that there are indeed intrinsic direction-dependent velocity differences between forward, backward, upward, and downward movements of the tongue dorsum, between adduction and abduction of the vocal folds, between raising and lowering movements of the velum, and between raising and lowering movements of the whole larynx. For example, velum raising is on average 18 % faster than velum lowering, and vertical tongue movements were found to be about 23 % faster than horizontal movements (all other factors being equal). Factors like sex or speaking rate were found to have an effect on the observed velocity differences, where particularly the effect of the speaking rate was unexpected. The findings are a first step towards a macroscopic biomechanical characterization of articulators and were discussed in the context of the complex interplay of neural control, the biomechanical system, and aerodynamic forces during the production of speech sounds. The results may have an impact on multiple theoretical and practical issues in speech research and technology, for example: • Considering the intrinsic velocities of articulators might improve articulatory speech synthesis. To this end, the differences were implemented into the articulatory speech synthesizer VocalTractLab for more realistic articulatory trajectories. The perceptual relevance of the more realistic trajectories with regard to the naturalness of the synthesis will be assessed in the near future. • Since we found that intrinsic velocities of articulators depend on biomechanical properties, they are also likely to vary with the individual physiology of speakers. Therefore, these properties might be worth exploiting in future speaker recognition and identification systems. • From the “embodiment” perspective of speech production, the expected results could contribute to the debate as to which biomechanical differences of articulators are exploited to shape speech sounds, and in which cases the differences are “equalized” by motor control. • The project results could help to understand asymmetric vowel distributions in the world’s languages better, as well as the sound changes that are found in many languages, because intrinsic biomechanical properties of articulators could explain the preferences of certain speech sounds over others in terms of articulation costs. • The results could significantly contribute to the dynamical perspective on speech production, which defines an intrinsic temporal metric for speech production based on the dynamical behavior of the articulators that is closely coupled to intrinsic velocities. • The data obtained in this project may be useful to verify biomechanical vocal tract or tongue models. These models have many parameters that define material properties of muscles and tissues, which are often hard to measure directly and need to be estimated. The characteristics of the articulators determined in this project could serve as a benchmark to verify the macroscopic behaviour of such biomechanical models.
Publications
- (2020). “An experimental study of tongue body loops in V1-V2-V1 sequences”. In: Journal of Phonetics 80, p. 100965
Thiele, Christian et al.
(See online at https://doi.org/10.1016/j.wocn.2020.100965)
