A semi-automatic method for extracting vocal tract movements from X-ray films

Abstract

Despite the development of new imaging techniques, existing X-ray data remain an appropriate tool to study speech production phenomena. However, to exploit these images, the shapes of the vocal tract articulators must first be extracted. This task, usually manually realized, is long and laborious. This paper describes a semi-automatic technique for facilitating the extraction of vocal tract contours from complete sequences of large existing cineradiographic databases in the context of continuous speech production. The proposed method efficiently combines the human expertise required for marking a small number of key images and an automatic indexing of the video data to infer dynamic 2D data. Manually acquired geometrical data are associated to each image of the sequence via a similarity measure based on the low frequency Discrete Cosine Transform (DCT) components of the images. Moreover to reduce the reconstruction error and improve the geometrical contour estimation, we perform post-processing treatments, such as a neighborhood averaging and a temporal filtering. The method is applied independently for each articulator (tongue, velum, lips, and mandible). Then the acquired contours are combined to reconstruct the movements of the entire vocal tract. We carry out evaluations, including comparisons with manual markings and with another semi-automatic method.

Region Segmentation in the Frequency Domain Applied to Upper Airway Real-Time Magnetic Resonance Images

“…However, compared to other currently used modalities such as X-ray [3], ultrasound [4],or electromagnetic articulography [5], MR imaging yields only moderate data rates. Due to the low spatio-temporal resolution of conventional magnetic resonance imaging (MRI) acquisition techniques, the earliest MR-based speech studies were limited to vocal productions with static postures such as vowel sounds (see [1] and the references therein).…”

“…At the same time, a new image processing challenge has been posed by the necessity of the contour extraction from the real-time MR images which are, generally speaking, of poor quality in terms of noise. A similar problem has been addressed in [3] for the case of X-ray image sequences showing the sagittal view of the human vocal tract. However, midsagittal MR images and sagittal X-ray images are quite different since the X-ray process only allows a projection through the volume of interest, i.e., the head of the subject, so that for instance, the teeth obstruct the view of the tongue.…”

High spatiotemporal cineMRI films using compressed sensing for acquiring articulatory data

“…For larger databases it is possible to take advantage of the fact that, unlike X-ray, MR imaging produce contours which do not overlap between each other. This feature is very interesting in terms of contour tracking and we exploited it by adapting the semi-automatic tracking algorithm developed by Fontecave and Berthommier [16]. The main principle is to delineate the target contour in key images randomly selected in the cineMRI, and then to index images where the contour has to be tracked with respect to the key images by using a DCT (Discrete Cosine Transform) distance.…”

Estimate articulatory MRI series from acoustic signal using deep architecture

“…For the training of the mapping models, rich articulatory datasets are needed, which should contain quantitative articulatory position data along with recordings of acoustic data produced. Various methods are used to record a speaker's articulatory movements, including Xray films [10,11], magnetic resonance imaging (MRI) series [12], 3D motion capture and electromagnetic articulograph (EMA). In the prior works, EMA data is the most widely used data for it has high temporal resolution.…”