One of the weaknesses of my dissertation research was that it was based on data from a cadaver, so it is difficult to relate the output of the model to the speech of a living person. The British Academy has generously provided funding so that I can address this issue. As of this writing, I'm in Edinburgh, at Queen Margaret University, working on this with Jim Scobbie.

On July 7, 2011, I spent some time supine in an MRI machine and got a lot of scans of my vocal tract. This is volumetric data, at very good resolution: the pixels of the images are 1.1875 mm square, and the images are spaced 1.2 mm apart. Each scan required about 20 seconds to acquire. I stabilized my vocal tract by phonating throughout the acquisition; I used creaky voice to make my breath last longer. We have (noisy) audio recordings, but I haven't looked at them yet.

I'm thrilled with the images, and can imagine them being useful to a variety of researchers. So, download and enjoy. My own publications from the images will (I hope) be forthcoming. If you do something and need to publish, send me an email and I'll fill in any gaps you need for the methods section.

The files below are compressed in7-zip compression format. The images are in DICOM image format. The standard image processing software isn't going to help you. You can view them with something like MicroDicom. Each of the compressed files below also has a MetaImage file, which a lot of medical-imaging-type programs can read.

If 3D data in DICOM format is a bit intense, I've also identified the sagittal slices and put them in a zip file here (~2 MB). They're in PNG format (which you computer will read), each having a file name like ɑ.png, ɪ.png, ʊ.png, etc. in order to identify the segment. These could be a nice teaching resource.

So, happy exploring. If you manage to do something intelligent with these images before I can, good on you!

Volumetric scans on my vocal tract

All contents copyright © 2018 Adam Baker, except where otherwise noted.