Siri is Apple’s virtual assistant, now available on numerous devices including the iPhone, Apple TV, and Apple Watch. Siri is easy to use, but many people wonder, How does Siri work? Quite frankly, it’s complicated. Siri was definitely not developed overnight, and decades of research and development went into perfecting Siri and other voice recognizing virtual assistants. Today, users can make Siri sing, search upcoming movies, and check the weather with merely their voice and a few seconds of patience. Here, we give a brief history of how this has been made possible and give an overview of the process that makes Siri work.
A Brief History
In the early days of computing (1950’s), voice recognition was already being developed. Programs had to be coded by hand for each syllable, so it’s no surprise that by 1961, a program by IBM could only recognize 16 words. By the 1980’s, the calculating power of computers increased significantly and more companies and countries pushed for voice recognition. Statistical modeling was now possible, which made it possible for computers to recognize strings of words to determine what the individual was saying.
While voice recognition had come a long way by the 1980’s, the 1990’s and onward has been critical for understanding that most people do not speak perfectly and clearly. Many people remember the days when speaking to a virtual voice through the telephone was a headache that required extreme patience and enunciation. Once out of the lab, speech recognition software has not stopped developing. Improvements have been made that consider the real world like the existence of background noise. Today, companies are pushing their software to learn more languages, determine the speaker’s mood, and engage in fluent two-way conversations.
The most important discovery for how Siri works is the development of speech recognition by IBM and numerous Japanese companies and universities was critical for developing the Siri we know today. Siri herself took decades and was first dreamed up when Apple developed the Knowledge Navigator in the late 1980’s, a theoretical avatar that users could interact with. While the avatar was only a hypothetical idea, it acted as a platform for developing the modern virtual assistant.
So How Does Siri Work?
Speaking a command to Siri involves numerous steps despite the nearly immediate response. When you say, “Search Mexican restaurants near me”, your device firsts encodes the sounds of your command. Next, the signal on your phone talks to your local cell tower, your internet service provider, and then a server in the cloud which contains the necessary tools to comprehend language.
Your device is also equipped to comprehend language, so at the same instant the cloud is processing your request, your device is as well. Your device also figures out if what you said needs help from the server in the cloud or if you command can be handled solely from the device. For instance, if you ask to open an app, your phone can handle that by itself. But to fulfill our request to search for Mexican restaurants, the device will need some extra help.
The server then analyzes your command, sound by sound, to determine what your request was. A type of statistical model comes into play since the server may come up with multiple combinations of words and sounds that you could have said. Usually the server or your device understand what you said, but sometimes users will get back the response “I didn’t quite get that” or Siri will ask for clarification.
Once the server reasonably determines what you were requesting, it will process the information, fulfill the request, and send the information back to your device for Siri to give you. All of this happens in just a few seconds, and with greater and greater accuracy each year. In the future, when people ask, “how does Siri work?”, the process may be different and Siri will likely be able to do much more like have fluent two way conversations. For now, we can marvel at the rich history of speech recognition and wonder what the future may hold.