Machines That Hear you Loud and Clear by Saskia Hewitt UWA Linguistics

Machines That Hear You Loud and Clear

If you’ve ever been frustrated by a machine on the end of a phone line that just couldn’t understand you, you’re not alone. Many a usually cool tempered person has slammed down their telephone to repetitions of, “I’m sorry, I didn’t hear you. Can you please repeat that?”

These types of machines are called speech recognition systems. They are often used by large companies in a customer service capacity, or for tasks like directory assistance.

But the problem, as any disgruntled telephone customer will tell you, is that they don’t always work as well as they should.

Dr. Roberto Togneri, from the Electrical Engineering Department in The University of Western Australia, is one of the researchers trying to develop speech recognition systems to a higher standard, so that they can be used more easily.

At the moment, he says, they are not very good at coping with ‘noise’, sounds that interfere with the signal that the machine picks up. Then, just like trying to hear someone talk in a noisy street, it becomes much harder for the machine to understand what is being said to it.

Dr. Togneri says his research is mainly focused on what happens to speech sounds when they hit the microphone and enter the machine. He works on the way machines process sound and on finding the features of speech that make it distinct from other types of sound.

“What makes speech speech?” he asks. “How do we eliminate effects from the environment, like noise?”

He says there also needs to be a mechanism in the machine to eliminate small differences in the way words are said by different people. That way, factors like accent would not make it so hard for the machine to register what words have been said.

Dr. Togneri says that the engineering aspects of speech recognition systems are just a small part of the whole. Researchers in other disciplines, like computer science, are also tackling problems such as ‘dialogue management’. This is what is going on when the machine asks you to repeat what you have said. It is able to realise when it has ‘misheard’ you, and it reacts so that it can fix the mistake. Computer scientists aim to make this process smoother and more efficient.

We will probably see speech recognition systems becoming more and more prevalent in our society as they become more sophisticated. They are already being used in dictation programs, and eventually researchers hope to add them to a range of electrical devices, such as televisions.

But even if we developed a perfect system, would people really want to use it? Or would they feel uncomfortable talking to a machine instead of a real person?

Dr. Togneri thinks that this may be true. He adds that in some cases, such as for dictation programs, it may always be simpler to use the traditional method (using a keyboard or touchtone type system) than voice recognition.

However, for people who find it hard to use these systems - for example, individuals with eyesight problems or a physical handicap - speech recognition systems can provide a way of connecting with technology that is easier and simpler than any other device.