Inter-Speaker Variability: Speaker Normalisation and Quantitative Estimation of Articulatory Invariants in Speech Production for French

Abstract

Speech production can be analysed in terms of universal articulatory-acoustic phonemic units shared between speakers. However, morphological differences between speakers and idiosyncratic articulatory strategies lead to large inter-speaker articulatory variability. Relationships between strategy and morphology have already been pinpointed in the literature. This study aims thus at generalising existing results on a larger database for the entire vocal tract (VT) and at quantifying phoneme-specific inter-speaker articulatory invariants. Midsagittal MRI of 11 French speakers for 62 vowels and consonants were recorded and VT contours manually edited. A procedure of normalisation of VT contours between speakers, based on the use of mean VT contours, led to an overall reduction of inter-speaker VT contours variance of 88%. On the opposite, the sagittal function (i.e. the transverse sagittal distance along the VT midline), which is the main determinant of the acoustic output, had an overall amplitude variance decrease of only 37%, suggesting that the speakers adapt their strategy to their morphology to achieve proper acoustic goals. Moreover, articulatory invariants were identified on the sagittal variance distribution along the VT as the regions with lower variability. These regions correspond to the classical places of articulation and are associated with higher acoustic sensitivity function levels.

Characterization of inter-speaker articulatory variability: A two-level multi-speaker modelling approach based on MRI data

“…Sec. I), including by Serrurier et al (2017) with the same data. The fact that no significant correlation (above 0.5) between MP2 and the strategy predictors and between S2 and the morphology predictors suggests that either speakers do not always adapt their strategies to their morphological characteristics or that these relations are not captured by the present linear modelling approach.…”

A multilinear tongue model derived from speech related MRI data of the human vocal tract

“…In the study by Stone et al (2016), the muscle architectures of different subjects were investigated. Speaker normalization was performed by Geng and Mooshammer (2009) and Serrurier et al (2017).…”

Speaker verification system based on articulatory information from ultrasound recordings

“…In this regard, articulatory information is known to be less affected by noise and contains speaker identity information, so it is prone to be used in automatic speaker recognition tasks [13]. Furthermore, several studies have shown articulatory related information is an important source of variability between speakers [13][14][15]. Nevertheless, measuring articulatory information is a complex task, and there are alternative strategies that have been proposed to estimate articulatory movements using acoustic information [16][17][18].…”