How does it work? | Speech recognition

Date:

2017-07-06 17:30:05

Views:

1365

Rating:

1Like 0Dislike

Share:

How does it work? | Speech recognition

The First device for speech recognition appeared in 1952, it was able to understand spoken human figures. 40 years later, the first commercial software for recognizing human speech. They were designed for people who, because of physiological characteristics could not type the text manually. Now the speech recognition is almost any smartphone, it allows us to interact with voice applications, facilitating and simplifying our lives. How does speech recognition — this was in today's issue.

Http://www.youtube.com/watch?v=PF6q8hUdKz8

If you speak a voice query, e.g., the destination address, the smartphone will not hear the street and house number, and the audio signal in which the sounds flow smoothly into each other without clear boundaries. The task of the speech recognition system — to restore the signal that has been said. It is worth noting that the same phrase pronounced by different people in different circumstances will be quite different from each other signals. To interpret them correctly makes the system acoustic modeling.

After giving a voice query, it is recorded by the smartphone and sent to the server, which determines the level of interference is samootdelku and the separation of the useful signal. Then the entry is divided into small pieces (frames), for example, with a length of 25 milliseconds in increments of 10 milliseconds, that is overlap. Thus one second of speech is a frameset.

First, each frame is passed through the acoustic model. System with machine learning, determine the variants of spoken words and context. The accuracy of the results depends on the completeness of the phonetic alphabet system. For each sound initially complex statistical model that describes the pronunciation of the sound in speech. The recognition system compares the incoming speech signal, phonemes, and from them collect the words. For example, the phonetic alphabet Yandex consists of 4000 elementary units that include phonemes and combinations. Each frame is mapped with more than one phoneme, but there are some that are suitable with varying degrees of probability. In addition, the system takes into account the transition probabilities, that is, determines which frames can follow a specific phoneme. For this purpose data on the pronunciation, morphology and semantics. Therefore, the system selects variants of words, which then analyzes the forms, parts of speech and possible statistical relationships between them.

Later in the process entering a language model with which the system determines the likely order of words and, if necessary, restores the unrecognized word in meaning based on the context and the available statistics.

As a result of information supplied in the main unit recognition system decoder. This software component combines the data from the acoustic and language models on the basis of their Association gives the final result in the form of the most likely sequence of words.

Thanks to the machine learning system is robust to noise and can recognize the speech with an accent. The accuracy of modern systems of speech recognition exceeds 90 percent.

Recommended

An air leak site has been found on the ISS. What's next?

An air leak site has been found on the ISS. What's next?

Air leak occurs in Russian station module Inside the International Space Station live astronauts from different countries and all of them need oxygen. The air needed for the life of the crew is produced by special equipment, but the tightness of the ...

Why can thinking about death make life happier?

Why can thinking about death make life happier?

Awareness of one's own mortality can be a liberating and awakening experience How do you feel about the idea of death? How often do you think about it and what emotions do you feel? Many of us have been pondering these questions lately. The pandemic ...

A new photo of Jupiter has found a new spot. What's it?

A new photo of Jupiter has found a new spot. What's it?

New photo of Jupiter taken by the Hubble Telescope Jupiter is considered the largest planet in the solar system. It mainly consists of a huge amount of hydrogen and helium, so it has a much lower density than many other planets. Most of all, Jupiter ...

Comments (0)

This article has no comment, be the first!

Add comment

Related News

How does it work? | Iris scanner

How does it work? | Iris scanner

the Technology of scanning an iris of the eye was first proposed in 1936 by ophthalmologist Frank Bursh. He said that the iris of each person is unique. The probability of coincidence is about 10 to the minus 78 degrees, which is ...

How does it work? | Fingerprint scanner

How does it work? | Fingerprint scanner

identification of the fingerprint — one of the most reliable ways to confirm the identity of the person. On the accuracy of this method is second only to the retinal scan and DNA analysis. Fingerprint — it's nothing li...

How does it work? | Computer mouse

How does it work? | Computer mouse

History of computer mouse originates 9 December 1968, when it was presented at the exhibition of interactive devices in California. The patent for this gadget got Doug Engelbart 2 years later. The first computer, the set which inc...