Categorising vowel space shape using geometric morphometrics

My project Antomy, Acoustics and the Individual makes use of geometric morphometrics to characterise and analyse vocal tract shape. However, this powerful technique is not limited to tangible, physical shapes; it can be used to describe and analyse the shape of anything. One thing that is very familiar to phoneticians is the vowel space, which describes the relative position of an individual's vowels on a F1-F2(-F3...etc.) axis. The shape of an individual's vowel space may be relevant for speaker identification, but until now there has not been a robust way of categorising and comparing their shapes. In this project we explore the potential for geometric morphometrics to provide useful information about the vowel space for phoneticians.

Lombard speech in MRI noise

When speaking in noise, our voice doesn't just get louder - the pitch, formants, and other characteristics also change. The type and amount of change in the voice depends upon the type of noise. MRI scanners are very noisy, and are used a lot for speech research; however no studies had been done to find out what effect the MRI noise has on speech despite it having very unusual spectral and temporal characteristics. We found that the MRI noise had a pronounced effect on the speech signal, even greater than the effect of lying down in the scanner (something which has been well studied). We therefore recommend that speech researchers take care to replicate MRI-like conditions if recording contemporaneous speech alongside MRI, and highlight the Lombard effect as an additional source of articulatory variation between the MRI data and normal speech.

Read our paper on this topic here (paper ID 303).

Active noise control

My dissertation project for my MSc in Digital Signal Processing focused on active noise control, and I developed a new algorithm making use of filtered-x and modified filtered-x structures. By incorporating a sparsity constraint into the algorithm derivation, we were able to improve the convergence speed of the algorithms.

Read our paper on this topic here.