Öhman returns: New horizons in the collection and analysis of imaging data in speech production research

There have been enormous technical advances in the use of imaging techniques in speech production research in terms of resolution and frame rates. However, a major bottleneck lies in the lack of appropriate data reduction and quantification methods which allow for a parsimonious representation of the high-dimensional image data. Particularly the rapid increase in frame rates seen in data acquisition makes traditional, error-prone methods of contour tracking unwieldy due to the high amount of manual intervention required. We discuss recent developments in methods that obviate contour tracking but instead process the entire image. Specifically, we focus on one such approach by demonstrating the application of Principal Component Analysis to ultrasound images. This method not only exploits the information present in the entire image, but it also straightforwardly allows for the representation of the temporal evolution of an utterance by reducing a series of images to a time-varying single-value representation. In an illustrative example, inspired by the seminal work of Öhman (1966), we show how characteristic patterns of coarticulation between vowels and consonants can be captured in this way.