Speech Changes May Detect Fatigue

 

Cynthia M.  LaJambe - cml149@psu.edu

The Thomas D. Larson Pennsylvania Transportation Institute

201 Transportation Research Building

Penn State University

University Park, PA 16802

Frederick M. Brown - f3b@psu.edu

Department of Psychology

Penn State University

University Park, PA 16802

Rebecca M. Reichardt - rebecca.reichardt@us.army.mil

Department of Psychology

Towson University

Towson, MD 21252

Malcolm Brenner - brennem@ntsb.gov

National Transportation Safety Board

Washington, DC 20594

Robert A. Prosek - rap6@psu.edu

Department of Communication Sciences and Disorders

Penn State University

University Park, PA 16802

 

Popular version of paper 4pSCa2

Presented Thursday afternoon, April 22, 2010

159th ASA Meeting, Baltimore, MD

 

 

Has anyone ever told you, You sound tired? More people are tired now than in generations past. According to the National Sleep Foundation, the average American adult gets 6.9 hours sleep per night as compared to the suggested 8 hours or more of sleep. Lack of sleep has serious consequences. It is estimated that fatigue due to sleep loss is the largest known cause of transportation accidents - accounting for at least 15 20% of accidents, more than alcohol and drugs. As reduced hours of sleep per night become the norm, accidents are likely to increase as alertness levels decrease. Fatigue monitoring technologies are urgently needed especially in transportation, military operations, and security industries where accidents can result in serious threats to public safety. We have chosen to study voice indicators of fatigue because often speech can be easily and quickly collected in work settings with minimal interference during hands-on duties. Our findings suggest that speech changes may detect the presence of fatigue.

 

During a recent sleep deprivation study we recorded speech from a task requiring counting as quickly as possible from 90 - 99 and reading from the standard Rainbow passage. Participants were 26 native English-speaking 18-26 year-olds in good health who slept a minimum of 7.5 hours or more per night. Vocal measures were compared between 13 speakers who were sleep deprived (SD group) for 36 hours and 13 non-sleep-deprived control group (CTRL group) participants. Speech was recorded during an evening baseline testing session at 8:30 pm on Day 1, and also at 8:30 am and 8:30 pm on Day 2 which for the sleep-deprived group was at 24 hours and 36 hours of sleep deprivation. From the speech data we are reporting here on measures related to fundamental frequency (pitch), intensity (loudness), reaction times, voice onset time or VOT (the length of time that occurs between the release of a stop consonant, i.e., b, p, d, t, g, k, and when vocal cord vibrations begin) from the Rainbow passage, and Formant 1-4 frequencies (spectral peaks of the sound spectrum) from the word three spoken during the Counting task. The vocal measures were examined for differences over time between the SD and CTRL groups which would indicate the presence of accumulating sleep deprivation effects in speech.

 

Our results showed that speech differences between SD and CTRL groups depended on the speech task and vocal measure. For the SD group, speaking rate was slightly less for the Rainbow passage (but not Counting) as sleep deprivation increased, and reaction time to begin speaking was much slower for Counting (but not the Rainbow passage). In contrast, CTRL speakers became faster at these tasks. Both the VOT for the stop consonant d and Formant 1 were reduced for SD speakers but increased for CTRL speakers over time. Measures for fundamental frequency and intensity were not different for the two groups on either task over time. These variables apparently lacked sensitivity to fatigue from increasing sleep loss.

 

The speech changes we found due to sleep deprivation are both similar to and different from other psychological and physical states that alter speech such as alcohol intoxication and hypoxia or lack of oxygen at high altitude (see Table 1 below).

 

Table 1. Some speech findings comparing psycho-physiological states to sleep deprivation.

 

	Response Speed	Speaking Rate	/d/	Formant 1
ACTIVATION
Stress/Workload			--
DEACTIVATION
Depression			No Change	No Change
Hypoxia			No Change	No Change
Alcohol Intoxication				No Change
Sleep Deprivation

 

The important point here is that a different pattern of speech changes occur with sleep deprivation as compared to speech changes resulting from other states of high and low activation. The ultimate goal is to find a vocal fingerprint that identifies fatigue in speech and rules out other psycho-physiological states as causes for speech alterations. Then instead of saying, You sound tired, well say, You ARE sleepy dont drive!