Measuring glottal activity during voiced speech using a tuned electromagnetic resonating collar sensor

Non-acoustic speech sensors can be employed to obtain measurements of one or more aspects of the speech production process, such as glottal activity, even in the presence of background noise. These sensors have a long history of clinical applications and have also recently been applied to the problem of denoising speech signals recorded in acoustically noisy environments (Ng et al 2000 Proc. Int. Conf. on Acoustics, Speech, and Signal Processing (ICASSP) (Istanbul, Turkey) vol 1, pp 229–32). Recently, researchers developed a new non-acoustic speech sensor based primarily on a tuned electromagnetic resonator collar (TERC) (Brown et al 2004 Meas. Sci. Technol. 15 1291). The TERC sensor measures glottal activity by sensing small changes in the dielectric properties of the glottis that result from voiced speech. This paper builds on the seminal work in Brown et al (2004). The primary contributions of this paper are (i) a description of a new single-mode TERC sensor design addressing the comfort and complexity issues of the original sensor, (ii) a complete description of new external interface systems used to obtain long-duration recordings from the TERC sensor and (iii) more extensive experimental results and analysis for the single-mode TERC sensor including spectrograms of speech containing both voiced and unvoiced speech segments in quiet and acoustically noisy environments. The experimental results demonstrate that the single-mode TERC sensor is able to detect glottal activity up to the fourth harmonic and is also insensitive to acoustic background noise.