Speech Testing Tools for Teachers

Professor Abeer Alwan, Director
The Speech Processing and Auditory Perception Laboratory

Preschool and elementary teachers at several Los Angeles area schools may soon have more time available for teaching, while still meeting national and state-mandated educational priorities.Experts in engineering and education at UCLA and other universities are developing a child-friendly testing system that measures and analyzes children's reading and pronunciation skills over time. They are working closely with elementary school teachers in the community to design an effective assessment system. Initial efforts will focus on five- to eight-year-old native speakers of American English and non-native speakers of Mexican backgrounds.

"To date, no one has tracked the development of early speech in this group of non-native English speakers," explained Alwan, a member of the UCLA Henry Samueli School of Engineering and Applied Science faculty. "And yet more than 60 percent of the students in some Los Angeles schools are native speakers of Spanish."

The Technologically Based Assessment of Language and Literacy project (or TBALL) will allow researchers to tackle several fundamental research issues. For instance, the project's acoustic and pronunciation modeling algorithms must address not only variability in speech from child to child, but also for a single child over time. It is also more difficult for computers to recognize children's speech because of their different acoustic characteristics, including higher pitch and resonant frequencies. The team is devising age-appropriate ways of displaying information to capture a child's attention and elicit responses. They are also determining what criteria are appropriate to use in scoring the children.

TBALL builds on a rich history of collaborative activities between engineering and education at UCLA, including improved curriculum feedback mechanisms, the Smart Kindergarten project, and the UCLA/LAUSD Computer Science Institute. The project will draw on the expertise of researchers in electrical engineering, computer science, and education at UCLA; electrical engineering, linguistics and neuroscience at the University of Southern California; and education at UC Berkeley.

"Currently, a teacher has a series of flash cards or a testing sheet and shows the child an image or word and asks him or her to say it aloud," explained Alwan. "Then he manually scores the child's pronunciation. There is little consistency in scoring from teacher to teacher, and the test offers only minimal assessment of a child's skill level." The TBALL team hopes to create consistency in scoring pronunciation, fluency and comprehension levels across classrooms through their automated system. Information derived from the project will also aid educators in determining which teaching methods are most effective with children of varying backgrounds.

The TBALL team will receive more than $3 million under a collaborative research grant from the National Science Foundation to support their research.