Until now, voice-recognition systems haven’t caught on widely among schools. Too many potential problems–background noise, variations in human speech, prohibitive cost–have kept them out of the education mainstream. But a program developed by researchers at Carnegie Mellon University aims to change that.

Carnegie Mellon’s Project LISTEN, which stands for “Literacy Innovation that Speech Technology Enables,” has developed an automated reading tutor that listens to children reading aloud, corrects their mistakes, and helps them develop literacy skills.

The project’s director, Jack Mostow, sees an enormous potential for the system in schools. “An automated reading tutor that listens could give millions of children individualized reading assistance that teachers can’t provide,” he said.

The program lets students choose from a menu of stories. Wearing a headset microphone, students read the text of the story aloud into the microphone as it pops up on the screen. The program intervenes when a student asks for help, makes a mistake, or gets stuck. At that point, the program “reads” the passage aloud for the student.

The reading tutor adapts Carnegie Mellon’s state-of-the-art Sphinx II speech recognition technology to analyze the students’ oral reading. Because the exercise is tightly structured, Mostow said, the reading tutor has very little trouble understanding the students’ speech.

The program runs under Windows 95 or NT 4.0 on a 200 MHz Pentium machine with at least 64 MB of memory. Headset microphones are used to cut down on interference from background noise and also let students use the program in a normal classroom setting, Mostow said.

In a 1996-97 pilot project involving third graders at Fort Pitt Elementary School in Pittsburgh, the program exceeded the school’s expectations. Six third graders who started about three years below grade level averaged two years’ progress in just under eight months.

“We were at the point where nothing seemed to work with these kids,” said Gayle Griffin, Fort Pitt’s principal.

But Project LISTEN gave the students enough repetitions with individualized instruction to allow them to develop their reading skills significantly, Griffin said.

The program also gave students who were self-conscious of their reading skills the chance to read aloud to a machine, Griffin said, which seemed less threatening to them than reading to a human tutor.

Last year, Fort Pitt expanded the program by installing machines in eight different classrooms. In January, a controlled study was conducted at the school under regular classroom conditions. One-third of the students tested used the automated reading tutor, a third used commercial software without voice input capabilities, and a third received regular classroom instruction.

Students using the voice-recognition system showed more gain in reading skills than the other groups, Mostow reported. But what was especially surprising, said Mostow, was that these students showed dramatic gains in reading comprehension as well as word recognition.

This fall, more than 100 Fort Pitt students in grades 2-5 will be using the automated reading tutor.

The program is still in a prototype stage, Mostow said, but Carnegie Mellon is working on licensing the technology for commercial development.

Carnegie Mellon University

Project LISTEN