What makes the things you say recognizable as coming from you?
There is a growing amount of research on personal voice quality but there remains little known about what actually characterizes a sound as coming from an individual speaker. To better understand what makes your voice yours, Jody Kreiman and Patricia Keating from the University of California, Los Angeles have banded together to carry out linguistics research applying acoustic tools.
Scheduled to report their findings at the 172nd Meeting of the Acoustical Society of America, Kreiman and Keating wanted to find out how to quantify how people sound like. After all, a number of factors can affect a person's voice over time, including an individual's health, emotional state and conversation context. If something was changed physically, for instance, they wanted to know how to predict the resulting sound.
"Voice quality is going to wander. We are looking at the point when you stop sounding like yourself and start sounding like someone else," said Keating.
Studying Voice Quality
According to earlier neuropsychology, cognitive psychology and phonetics studies, a protoype is built for every talker based on what listeners organize as intra-talker variability. However, this average representation of a voice comes with a set of deviations. Even just one syllable can contain distinguishing information for a voice but it's not clear which characteristics specifically act as the most important identifiers or how much of these characteristics must differ before a voice is deemed unrecognizable.
For the study, the researchers carried out digital analysis on recordings made of 50 native English-speaking women. The participants were made to read five sentences two times across three different days.
The researchers looked at acoustic parameters for consonant and vowel sounds in the read sentences, like intensities in harmonic frequencies in relation to each other and fundamental frequency, and how these parameters compare to underlying levels of noise in the voice.
With the read sentences providing characteristics with quantitative range and average, a semblance of an identifying voice profile was formed. The researchers compared the participants' profiles, testing them against a set of characteristics to see if the correct speaker can be accurately distinguished.
Future Voice Research
Kreiman and Keating had previously worked on similar research but were involved with a larger sample size this time around, which offered them more insight into what makes a voice recognizable.
The results of their current study will help in answering questions like what's defined as confusing in voices in the context of both human and computer listeners.
