All the secrets of artificial intelligence after the "Hey Siri" -


Post Top Ad

Post Top Ad

Thursday, 19 October 2017

All the secrets of artificial intelligence after the "Hey Siri"

Apple just left us jaw-dropping with the publication in his Machine Learning Journal explaining how Siri's voice activation works, come on, the classic "Hey Siri." What happens after these two magical words is simply fascinating .

Apple explains that the iPhone and Apple Watch microphones transform our voice into a waveband with a ratio of 16,000 undulations per second before the detector decides that you are trying to summon Siri with your voice:

    A spectrum analysis converts the wave sample into a sequence describing the sound spectrum at about 0.01 seconds. Approximately 0.2 seconds of audio is enough for the neural network to convert those patterns into a probable distribution with a series of sound parameters, specifically those used in the phrase "Hey Siri" + a silence + order.

In addition, it has a variable to decide if you are invoking Siri or not :

    We have some flexibility to facilitate the activation of Siri in complicated situations where it is easy to have false activations. There is a normal limit and a lower one to activate Siri, which normally drops the activation. If a phrase enters the lower and normal limits, the system enters a sensitive phase because it fears having lost an invocation, so that if the user repeats the phrase, even with the same background noise and volume, Siri is activated. This second mechanism improves the usability of Siri significantly without the need for the false alarm to increase significantly.

We know that "Oye Siri" relies on the co-processor of the iPhone to hear the activation word without the need for physical interaction or decrease the life of the battery. However, "hey Siri" works differently on the Apple Watch , since it requires the display turned on. Of course, as much uses the 5% processing

    The "Hear Siri" detector only works when the watch's coprocessor detects a wrist gesture, which turns the display on. At that point, WatchOS is ready to operate, prepare the screen, etc. "Hey Siri" only uses 5% processing capacity. It is a challenge to start capturing audio in time to catch the activation phrase, so we grant permission to start the detector as soon as possible.

Why did you choose the phrase "Hey Siri" as an activation?

That's a good question, and maybe I could have opted for a longer one . In Apple they explain it like this:

    Before the "Hey Siri" function existed, a small portion of users using Siri would start their sentences by pressing the Home button. So we started with "Hey Siri" in the initial training in English. We used other examples as well, but this gave us more precision. Also, we were careful with the pronunciation of Siri, so that it was not confused with other similar words.

We imagine that in Castilian, many people also used "Oye" as initial entry , so the choice was simple.

Via | 9to5Mac 

No comments:

Post a Comment

Post Top Ad