This speech recognition software is much faster than human typists

Speech recognition software used to be awful. It couldn't deal with background noise, its vocabulary was limited and it was awkward to use in public.

But we've come a long way. A new speech recognition system can transcribe English or Mandarin about three times faster than humans can type on a smartphone, according to a recent study. Developers from the popular Chinese search engine Baidu created the program back in December. It's called Deep Speech 2, and it uses machine learning to vastly improve speech recognition.

The study, a collaboration between Stanford University, Baidu and the University of Washington, also found that the system produced 20.4 percent fewer errors than people typing in English and 63.8 percent fewer than people working in Mandarin.

In the experiment, 32 individuals would either type on a QWERTY smartphone keyboard or speak sample phrases such as "physics and chemistry are hard," "have a good weekend" and "go out for some pizza and beer."

To make the conditions as authentic as possible, the human subjects were also allowed to use autocorrect and word suggestions. And all participants were between 19 and 32 years old.

"We’re putting speech recognition up against people who are really good at this task," study co-author James Landay told Stanford News.

Overall, participants preferred the speech recognition system, though they had some reservations.

"When I made mistakes it seemed like it took longer to correct, because I was switching from holding the phone to speak vs. holding to type," said one participant, according to the study.

Another participant admitted to simply being more comfortable on their keyboard. "I could correct errors as they were made," the participant said "That being said, it seemed like I introduced more errors typing that the speech system did."

The researchers suggested some improvements for future versions of the system. For example, it would be great if the system could auto-detect when a speaker was finished instead of pressing a button. The researchers would also like to test it in a variety of noise levels.

Baidu doesn't plan to make the software available to the public, but they are integrating it into Baidu apps in China.

Even so, Landay sees the potential for widespread use of this speech recognition system.

“You could imagine an interface where you use speech to start and then it switches to a graphical interface that you can touch and control with your finger,” Landay said. “We should put speech in more applications than just typing an email or text message."

Watch an in-depth explanation of the experiment below: