In this study, 40 digital voice recordings were obtained from subjects recruited from the Montreal General Hospital. Subjects were either smokers of the equivalent of 1 package of cigarettes daily for at least 20 years, or were lifetime non-smokers. The audio file recordings were then evaluated by three groups: three adult voice experts, two adult non-experts, and two minors. Two voice doctors and a speech pathologist were the experts in this study. The test subjects were then classified according to gender and whether they were a smoker or not.
The results of these evaluations were surprising. It was expected that the voice experts, with their training and experience, would have the best or most accurate results. The non-expert minors, however, were correct 66% of the time in distinguishing smokers or non-smokers. This was the highest score in the experiment; it was higher than both the experts and non-expert adults. The minors were also quite proficient at detecting the gender of the person (97% overall). However, the errors that were made in determining gender occurred with female voices with a lower than expected fundamental frequency. This change in fundamental frequency has been previously described in smokers.
The non-expert adults averaged a score of 60%. That is the second best score, after the minors and before the experts. The adult non-experts determined the gender of the person correctly 99% of the time.
The experts determined if the person was a smoker or not 51% of the time. That is significantly less then the adult non-smokers and the minors. There are a few possible reasons for this. I think that maybe the experts tried too hard at evaluating the smoker’s voice scientifically, while the non-experts and minors went with their gut impression. In other words, the experts may have ‘over-evaluated’ and ‘over-analyzed’ what they heard, thinking this would increase accuracy. The experts were also under pressure as they were expected to get all the answers correct and this may have been a factor. It is also possible that in practice voice experts use and rely on visual and olfactory clues such as wrinkles, nicotine stains and the smell of stale smoke to help decide the smoking status of an individual. Adult non-experts and minors may be less aware of these clues and therefore less influenced by them. Minors may be listening ‘differently’ and have fewer inherent biases in their evaluations.
Interestingly, all three groups of evaluators found it harder to correctly assess the smoking status of female voices. This was particularly true for the expert group. The experts correctly identified the smoking status in 48% of females and 64% of males. The non-experts correctly identified the smoking status in 61% of females and 66% of males. Minor evaluators correctly identified the smoking status in 63% of females and 72% of males. These results suggest that there may be some characteristics of the male voice which make it easier to distinguish between a smoker and non-smoker.
My hypothesis was that the three groups of evaluators consisting of voice experts, adult non-experts and minors could determine smokers from non-smokers based on a voice evaluation alone. It was also expected that experts would have the best results. My hypothesis was incorrect in that regard. Minors performed best on all smoking versus non-smoking evaluations with an overall accuracy of 66% compared to 51% for the experts. Based on these results, determining smoking status based on voice alone appears to be difficult for all evaluators, with voice experts having results no better than chance alone. Minors seem to be better able to determine smoking status for both men and women, with results of 72% and 63% respectively. Although smoking is known to permanently lower fundamental frequency in heavy smokers, the difference may not be great enough to allow a consistently accurate determination of smoking status in a large majority of individuals.
These early results are interesting but would have to be verified by a larger study. The number of evaluators and test subjects was rather small and larger numbers may have shown different results. Another potential study could involve determining smoking status based on audio files alone versus audio files with visual clues to see if this makes a difference. The person conducting the experiment would see whether sensory clues help evaluators get a better score.