Digital Speech Analysis Tests Sobriety

May 23, 1996

Slurred speech is often a sure sign that someone's been drinking.

Now, a Georgia Institute of Technology researcher is working with colleagues from Indiana University to digitally quantify this telltale sign, which could lead to a simple, non-invasive way to test a person's sobriety.

"This is basically an effect of fine motor control," said Kathleen E. Cummings, a lecturer in Georgia Tech's School of Electrical and Computer Engineering. "We're looking at specifically what happens during speech production at your vocal cords, how steadily you can produce the excitation (air from your lungs) going through your vocal cords."

Preliminary results show that intoxicated speech is marked by jumpy changes in pitch and energy production and unsteady opening and closing of the vocal cords.

Cummings discussed her work May 16 at the 131st annual meeting of the Acoustical Society of America in Indianapolis. She is working with Dr. David B. Pisoni and Dr. Steven B. Chin of Indiana University, as part of their ongoing study of ways to measure how alcohol consumption affects speech. The current project is sponsored by the Alcoholic Beverage Medical Research Foundation.

Pisoni, director of Indiana's Speech Research Laboratory and a professor of psychology, is considered a leader in the study of acoustical analysis, synthesis and perception of speech. Chin is a psychology postdoctoral student specializing in linguistics.

The two researchers approached Cummings after hearing about her thesis work at Georgia Tech, published in 1992, on how speech changes when produced under emotional stress or with linguistic effects such as talking quickly or slowly, loudly or softly.

"Given her robust results in the differentiation of styles of stressed speech, we thought that this type of analysis might show characteristic changes in speech produced under alcohol," Chin said.

For her thesis work, Cummings used digitized speech collected from several people speaking in 11 of the most common non-normal styles of speech. She then spent several years analyzing the signals produced by the sounds, looking specifically at the glottal excitation waveform.

During speech production, air passes from the lungs through the glottis, an opening in the vocal cords, then is shaped into sounds by parts of the vocal tract, such as the teeth, tongue and lips. If the glottis stays open, the result is unvoiced sounds like "p" and "t." If it opens and closes periodically, voiced sounds, like "b," "z" and vowels are produced.

The glottal excitation waveform is the puffs of air produced by the opening and closing of the glottis during voiced speech. Cummings concentrated on voiced sounds in order to study the glottal excitation waveform, which is known to be important in the subtle parts of natural speech, such as emotion and style.

She discovered distinct differences between normal speech and that produced under emotional stress, with an accuracy rate of over 90 percent.

For her current research, Cummings said, "the idea is, can we do the same thing with sober versus intoxicated speech? If we have a sample of somebody's speech from an accident or at a particular time, can we analyze it and say, 'Yes, this person is intoxicated,' if we compare this to his normal, sober speech sample?"

To find out, Pisoni and Chin sent Cummings samples of sober and intoxicated speech from four different people, gathered at Indiana University. They include different types of speech, such as monosyllabic words, tongue twisters, isolated sentences and passages of connected sentences.

Samples were taken when participants were sober, moderately intoxicated (.05 percent blood alcohol level) and highly intoxicated (.10 percent blood alcohol level or higher, considered legally drunk in most states).

Past perceptual research on this database has shown that a person listening to the samples can reliably discriminate between sober and intoxicated speech. Acoustic analysis also has shown that intoxicated speech is slower, features longer sentences and is marked by mispronunciations, such as slurred sounds and transposed letters and words.

In the current study, Cummings is finding that alcohol has a major effect on the excitation parameters that reflect the steadiness with which a person produces speech.

Four speaker samples may not sound like enough for a comprehensive study, but Cummings said they form a sufficient database to make generalizations.

"If you see consistently the same trend between sober and intoxicated speech for four different speakers, that's actually a lot," she said.

Also, Cummings plans to continue her research on the other five speakers in the Indiana database.

Although much work is left to be done, Cummings said translating her research into a practical public safety device could be relatively easy. Law enforcement officials could record someone's speech at an accident or traffic stop, then analyze it later against a sample taken at a different time.

"If I can come up with a small set of parameters that differentiate sober and intoxicated speech, which I think I can do, it's actually not a hard task," she said. "There are some really simple distance measures that involve very few calculations."

The analysis would be done by computer, based on a mathematical formula that would yield a percentage probability as to whether the speaker was intoxicated.

The only stumbling blocks could be recording quality and legal issues, such as a person's refusal to give two samples for comparison.

"Now, if you ever got really, really lucky, and you found something that you only ever saw in intoxicated speech ... then you would be able to just do it on the fly," Cummings said. "But I haven't seen anything like that yet."

More importantly, researchers have to compare their results against other factors that alter the way a person speaks, such as speech impediments, injuries, diseases or even common colds. Ataxic dysarthria, for example, is a neurological condition that causes a person to sound intoxicated.

With more than a year's worth of work behind her and at least that much to go, Cummings hopes to soon isolate a distinct set of parameters that define intoxicated speech with a least 90 percent accuracy. Regardless, she and her colleagues hope their research adds to the basic knowledge and understanding of how speech is produced.
430 Tenth St. N.W., Suite N-112
Georgia Institute of Technology
Atlanta, Georgia 30318

John Toon (404-894-6986);
FAX: (404-894-6983)

Kathleen E. Cummings (404-894-3335); Internet:

WRITER: Amanda Crowell

Georgia Institute of Technology

Related Alcohol Articles from Brightsurf:

Alcohol use changed right after COVID-19 lockdown
One in four adults reported a change in alcohol use almost immediately after stay-at-home orders were issued: 14% reported drinking more alcohol and reported higher levels of stress and anxiety than those who did not drink and those whose use stayed the same.

Changes in hospitalizations for alcohol use disorder in US
Changes over nearly two decades in the rate of hospitalizations and in-hospital deaths from alcohol use disorder in the US were examined in this study.

Associations of alcohol consumption, alcohol-induced passing out with risk of dementia
The risk of future dementia associated with overall alcohol consumption and alcohol-induced loss of consciousness in a population of current drinkers was examined in this observational study with more than 131,000 adults.

New alcohol genes uncovered
Do you have what is known as problematic alcohol use?

Does estrogen influence alcohol use disorder?
A new study from researchers at the University of Illinois at Chicago shows that high estrogen levels may make alcohol more rewarding to female mice.

Sobering new data on drinking and driving: 15% of US alcohol-related motor vehicle fatalities involve alcohol under the legal limit
A new study in the American Journal of Preventive Medicine, published by Elsevier, found that motor vehicle crashes involving drivers with blood alcohol concentrations (BACs) below the legal limit of 0.08 percent accounted for 15% of alcohol-involved crash deaths in the United States.

Alcohol-induced deaths in US
National vital statistics data from 2000 to 2016 were used to examine how rates of alcohol-induced deaths (defined as those deaths due to alcohol consumption that could be avoided if alcohol weren't involved) have changed in the US and to compare the results by demographic groups including sex, race/ethnicity, age, socioeconomic status and geographic location.

Cuts in alcohol duty linked to 2000 more alcohol-related deaths in England
Government cuts to alcohol taxes have had dramatic consequences for public health, including nearly 2000 more alcohol-related deaths in England since 2012, according to new research from the University of Sheffield's School of Health and Related Research (ScHARR).

Integrated stepped alcohol treatment for people in HIV care improves both HIV & alcohol outcomes
Increasing the intensity of treatment for alcohol use disorder (AUD) over time improves alcohol-related outcomes among people with HIV, according to new clinical research supported by the National Institutes of Health.

The Lancet:Targets to reduce harmful alcohol use are likely to be missed as global alcohol intake increases
Increasing rates of alcohol use suggest that the world is not on track to achieve targets against harmful alcohol use, according to a study of 189 countries' alcohol intake between 1990-2017 and estimated intake up to 2030, published in The Lancet.

Read More: Alcohol News and Alcohol Current Events is a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for sites to earn advertising fees by advertising and linking to