Team: Lisa Gustavsson (PI), Ellen Marklund, Anna Ericsson
Funding: 90 000 kr from Magnus Bergvalls stiftelse (2018-02869)
Project time: spring 2019 – fall 2019

The present project evaluates an automated formant estimation procedure, which takes into account variation in vocal tract length and configuration resulting both from different speakers and vowels.

The procedure entails using common practice automated formant estimation techniques (e.g., linear predictive coding; LPC), but automatically adjusting the formant ceiling to the optimal setting for individual speakers and vowels. The procedure was introduced by Escudero and colleagues (2009) and has since been used by other researchers (Wang, Seidl & Cristia, 2015; Marklund & Gustavsson, in preparation). It has however not yet been thoroughly evaluated. The present project will do so by applying it to synthesized vowels with known formant values and comparing the results to formant estimations without applying the individually adjusted formant ceiling setting.

Since high fundamental frequency remains a problem for formant estimation in general, the fundamental frequency of the synthesized vowels will be varied in order to map out the scope of the evaluated procedure’s usefulness.