What is the sound of one person talking? New speech collection tells all, syllable by syllable

May 01, 2006

COLUMBUS , Ohio - When researchers interviewed 40 Columbus residents about their opinions on life in the city, the scientists ignored what the people had to say.

All the scientists really cared about was how they said it.

In a lab at Ohio State University , researchers recorded 30- to 60-minute casual interviews with these residents, talking about how Columbus has changed over the years, how families should get along, as well as issues in sports, traffic and politics.

The participants were told that the purpose of the study was to learn how people express "everyday" opinions in conversation. But when the interviews were over, the scientists didn't even examine the views that were expressed.

Instead, they painstakingly listened to every spoken word - indeed every syllable - along with coughs, laughs and pauses in the conversation, and then labeled what actually was spoken.

The result is a 306,652-word repository of how people in central Ohio speak. Now scientists from around the world, and in a variety of disciplines, can use the collection - called the Buckeye Speech Corpus - to advance their research. (Researchers refer to a collected body of recordings as a corpus.)

"A critical part of communication is how your ears translate the sounds coming out of my mouth into recognizable words," said Mark Pitt, one of the leaders of the project, and professor of psychology at Ohio State.

"You don't have to go to school to learn how to do it. It is fast and efficient. The scientific question is 'how is this done?' The corpus will help researchers find answers to that question and many others."

Other members of this large research team include Eric Fosler-Lussier, assistant professor of computer science and engineering, and Elizabeth Hume, professor of linguistics, both at Ohio State.

The corpus, available to researchers on DVD, includes the actual recordings of the interviews, a written transcript, and several levels of labeling. The first half of the corpus was officially released in March. A typical display of a stretch of speech from the DVD shows two visual representations of speech from one of the talkers, indicating volume and frequency. Below the displays is a line with a transcription of what the speaker is saying. Then comes a line of "phonemic labeling," which shows how each word spoken was pronounced - similar to the pronunciation guides found in a dictionary.

This phonemic labeling is a key part of the corpus, because in casual conversations, people don't always use the proper English pronunciations they were taught in school, said Laura Dilley, a post-doctoral researcher in psychology who works on the corpus.

"It's interesting because of the many kinds of modifications that happen to the acoustics of speech in conversations," Dilley said. "For example, we may pronounce 'don't you' as 'don'tcha' when we are talking.

"By looking at the phonemic transcriptions we can tell exactly how a person pronounced a word. And from that scientists may be able to make some inferences about why people produce the sound one way as opposed to another."

The interviews are also labeled to show when people made non-word sounds during a conversation, such as a cough.

"Even laughs are noted, because some researchers may be interested in how that is part of a conversation," Pitt said. "It can convey meaning as well."

The 40 residents were interviewed in 1999 and 2000. But it has taken until this year to label the speech of half of the speakers, Pitt said. The data from the remaining 20 participants should be available by the end of the year.

"Collecting the interviews took a long time. But what really takes a long time is labeling them. Researchers had to listen to each conversation and figure out what sounds of the language the speakers said or didn't say. It's very difficult and very time consuming," he said.

When you actually sit down to label each word, "they can blend together and be very difficult to identify," Pitt said. "Sometimes you scratch your head trying to figure out how to label the speech. But when you're just listening to it as part of the conversation, it is all perfectly intelligible. That's part of the mystery of communication that scientists are trying to figure out."

A combination of factors makes the Buckeye Speech Corpus a unique resource for researchers, according to Pitt and Dilley. For one, it is one of the largest corpora of high-fidelity, conversational speech available. Other corpora involve speakers reading words directly from a text, but that is very different than conversation. There are also corpora of conversational speech recorded over the telephone, but that does not have the high fidelity of this corpus, and the dynamic of face-to-face conversation.

Another key difference is that the Buckeye Corpus is available free to researchers, both in academia and industry.

Although the Buckeye Corpus has been available for only a few weeks, researchers from around the world are already ordering copies of it. A researcher from Italy ordered the corpus to study "slips of the tongue," such as when people accidentally substitute one word for another that sounds similar (saying "rabbit" instead of "habit").

Pitt and Dilley expect a lot more interest in the corpus.

"Across a variety of fields -- in communication, speech sciences, linguistics, speech and hearing, computer science, psychology - the corpus will provide different uses," Pitt said. "It was created for all these scientific communities."

Computer scientists could use the corpus to help improve speech recognition software. Communication researchers will find it useful to study conversational dynamics. Psychologists will be interested in what listeners are faced with in terms of the physical and acoustic properties of speech.

Eventually, the corpus will have a search function that will make it even more useful. "If a researcher wanted to look up all the uses or pronunciations of a certain word in the interviews, he or she could do that," Dilley said.

The work of transcribing and labeling the interviews allowed the Ohio State researchers to uncover some interesting facts about how people speak in conversation.

For instance, the speakers spoke a total of 306,652 words, but only 9,600 different words. Almost 80 percent of the total words spoken were one-syllable words. One and two-syllable words made up more than 90 percent of the total words spoken.

Slightly more than half of the words spoken (57 percent) were function words - words like prepositions, pronouns and conjunctions that have mostly grammatical uses within a sentence. The remaining 43 percent were content words, which includes nouns, verbs and adjectives.

The Buckeye Speech Corpus project was originally funded with a seed grant from Ohio State's Office of Research. It has since received funding from the National Institute on Deafness and Other Communication Disorders, which is part of the National Institutes of Health.

Ohio State may be emerging as a center for corpora of conversational speech, Pitt said, as the Department of Speech and Hearing Sciences was just awarded a large, multi-year grant from NIH to collect interviews with speakers native to Ohio, Wisconsin, and western North Carolina.
Researchers can learn more about the corpus, and order a copy, at http://buckeyecorpus.osu.edu

To hear an audio clip of a woman talking that is a part of the Buckeye Speech Corpus, please click here

Contact: Mark Pitt, (614) 292-4193; Pitt.2@osu.edu

Laura Dilley, Dilley.28@osu.edu

Written by Jeff Grabmeier, (614) 292-8457; Grabmeier.1@osu.edu

Ohio State University

Related Speech Articles from Brightsurf:

How speech propels pathogens
Speech and singing spread saliva droplets, a phenomenon that has attracted much attention in the current context of the Covid-19 pandemic.

How everyday speech could transmit viral droplets
High-speed imaging of an individual producing common speech sounds shows that the sudden burst of airflow produced from the articulation of consonants like /p/ or /b/ carry salivary and mucus droplets for at least a meter in front of a speaker.

Speech processing hierarchy in the dog brain
Dog brains, just as human brains, process speech hierarchically: intonations at lower, word meanings at higher stages, according to a new study by Hungarian researchers.

Computational model decodes speech by predicting it
UNIGE scientists developed a neuro-computer model which helps explain how the brain identifies syllables in natural speech.

Variability in natural speech is challenging for the dyslexic brain
A new study brings neural-level evidence that the continuous variation in natural speech makes the discrimination of phonemes challenging for adults suffering from developmental reading-deficit dyslexia.

How the brain controls our speech
Speaking requires both sides of the brain. Each hemisphere takes over a part of the complex task of forming sounds, modulating the voice and monitoring what has been said.

How important is speech in transmitting coronavirus?
Normal speech by individuals who are asymptomatic but infected with coronavirus may produce enough aerosolized particles to transmit the infection, according to aerosol scientists at UC Davis.

Using a cappella to explain speech and music specialization
Speech and music are two fundamentally human activities that are decoded in different brain hemispheres.

Speech could be older than we thought
The theory of the 'descended larynx' has stated that before speech can emerge, the larynx must be in a low position to produce differentiated vowels.

How the brain detects the rhythms of speech
Neuroscientists at UC San Francisco have discovered how the listening brain scans speech to break it down into syllables.

Read More: Speech News and Speech Current Events
Brightsurf.com 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 Amazon.com.