Speech Recognition Is Officially Faster Than Typing
Since the advent of typewriters, people have used the classic QWERTY keyboard as their communication gateway. More and more, however, the human voice is getting into the mix—you can now search the web, make phone calls, and compose messages without touching a single thing. According to a 2016 study from Stanford University, the University of Washington, and the Chinese internet company Baidu, voice power is becoming even more powerful. For the study, researchers pitted 32 human smartphone users—16 English speakers and 16 Mandarin Chinese speakers—against a Baidu program called Deep Speech 2. The human participants would speak and then type (or type and then speak) short phrases into an iPhone such as "elections bring out the best" and "wear a crown with many jewels." The researchers left out punctuation to even the playing field, but did allow for autocorrect to help ensure their results could translate to the real world. In both English and Chinese settings, the participants were able to compose the phrases about three times faster with speech than with typing. Speech was also more accurate, with a 20% lower error rate in English and a whopping 63% lower error rate in Chinese. This could mean big things for speech input in apps and other technology of the future, especially those for which speed is paramount. Explore the relationships between speech and tech with the videos below.
The Science Of Talking With Computers
How we use our voices to communicate with tech.
Check Out This Month's Most Popular Topic
Study Shows Speech Recognition Is Faster Than Typing
The software left the human thumbs in the dust.
from Stanford University
How Does Speech Recognition Work?
Explore the history and technology behind voice recognition.
Key Facts In This Video
Early forms of voice recognition software had very little vocabularies. (0:34)
Google uses an artificial neural network for voice recognition, similar to your brain. (2:12)
Privacy is a major issue in voice recognition software. (3:22)