Digital Life (Absolut01) - Chapter 9

[Nine.]

Fang Zhiqiang knew that the method he proposed was a relatively correct method.

In terms of speech recognition, recognition methods based purely on speech and structural grammar may be more suitable for languages such as English, French, German, because the grammatical structures of those languages have always been relatively complete, and there is a language research that has been repeatedly studied for hundreds of years. Results-based, easier to digitize in terms of speech recognition.

But Chinese is not. The Chinese system is too complicated. The rupture between classical Chinese and modern Chinese is far more severe than the difference between English Middle English and modern English.

In daily use, even modern people will unconsciously use some words, sentences and even grammar in ancient Chinese. Middle school students have the most headache in the process of learning ancient Chinese. In daily use, it is everywhere. What should I do? Can a relatively complete grammatical structure be established to match the processing after speech recognition?

Starting from semantics and pragmatics, with intelligent programs as the core and common sense judgment as the basis, it should be possible to solve the problem of machine recognition in Chinese.

However, this is also an almost impossible task.

Semantics and pragmatics are not content that can be simply systematized, but an ever-changing system with inherent laws to follow.

I don’t know how many linguists have studied semantics and pragmatics in China, and it seems that there have been no major achievements over the years. So, can Lu Zhenyu make his own achievements?

Although, with the changes in the university system, most of those linguists are fooling people who write papers to mix up their qualifications, but the basic knowledge is still very solid, and Lu Zhenyu can be said to know nothing in this regard. With basic linguistics tutorials that you can easily find on the market, it’s not enough anyway.

Fang Zhiqiang said, “Xiao Lu, give me an address, and I’ll send you some materials later.”

“Well, then thank you uncle.”

After chatting for a while, Lu Zhenyu hurried to leave.

Fang Zhiqiang’s method may be a big problem for others, but for Lu Zhenyu, who already has Xiaoyu, it is not a big problem.

Xiaoyu’s intelligence has fully understood the complex language environment, and the judgment of semantics and pragmatics is not a big problem for Xiaoyu, and the experience of hanging on the Internet and continuously soaking in Warcraft has allowed Xiaoyu to learn a lot of comparisons in modern Chinese. Special expressions, especially some languages that are popular among young people.

According to Fang Zhiqiang, what Lu Zhenyu has to do is to directly connect the voice system to Xiaoyu’s current platform. Although the existing voice recognition system is not perfect, the general framework is good, and Xiaoyu’s independent judgment and learning will Quickly improve the recognition ability of the entire speech system.

“Xiaoyu, I split the front end of this speech recognition software. After the speech is input, it will automatically output the data of the Chinese characters corresponding to the pronunciation, but it will no longer be automatically selected. What word to choose, how to combine words and sentences, how to punctuate, these are all It’s up to you.”

“Understood.” After explaining the principle to Xiaoyu, Xiaoyu agreed to Lu Zhenyu’s plan after nearly 4 hours of independent calculations.

According to Lu Zhenyu’s proposal, Xiaoyu also separated the modules that may be used in speech recognition, stripped off the accumulation of common sense that he had accumulated for a long time, and only kept the basic communication environment, and divided the professional knowledge modules.

Therefore, although Xiaoyu himself is still leading the voice system for this trial run, it already has the basic structure of an ordinary voice recognition system.

For Xiaoyu, this is meaningless, but for Lu Zhenyu, as long as it is proved that such a basic framework is feasible, then the program modules based on such a basic framework can be gradually realized. Only the most core intelligent discrimination system, Some functions of Xiaoba may be used.

“Trial listening, first time: one, two, three.” After hooking up, Xiao Yu prompted to start the audition.

“Listen, the first result: one, two, three.” Xiao Yu’s interface dialog box showed the correct result, although it was too simple.

“Audio, the second time: Autumn is here, the weather is cold, and a flock of geese fly south.”

“Audio, the second result: Autumn is here, the weather is cold, and a flock of geese fly south.” The result is still correct, Lu Zhenyu Invigorated a little.

“Audition, the third time: Mercy is not forced, it is like a rain of rain that falls from heaven on earth; it is blessed not only for those who give, but also for those who give; it has above all innocence The power of the emperor is more than a crown to show the nobility of an emperor: the royal scepter only symbolizes the authority of the world, making the people fear the dignity of the emperor; the power of mercy is higher than the power, and it is deeply hidden in the emperor’s dignity. The heart is a kind of virtue that belongs to God. If the law enforcement person can adjust the mercy and justice, the power in the world is no different from the divine power of God. Therefore, Jew, although what you ask for is justice, please think about it. I think, if the rewards and punishments are really executed according to justice, there is no hope for anyone to be saved after death; since we pray for the mercy of God, we should do some merciful deeds according to the instructions of our prayers. I said this. , in the hope that you will be able to make some concessions from your legal standpoint; but if you insist on the original request, the courts of Venice are disinterested and have to convict the merchant.” (Quoted in ” The Merchant of Venice”)

This time, after a long time of calculation, Xiao Yu showed the result. Although the word judgment was correct, the punctuation marks were wrong a lot.

However, this was already much better than Lu Zhenyu expected.

After more than half a day and 400 rounds of testing and running in, Xiao Yu was able to hear Lu Zhenyu’s words indiscriminately, and even some obscure expressions and words couldn’t help it.

After all, Xiaoyu’s current knowledge is vast, far exceeding that of ordinary humans.

Although Lv Zhenyu is currently using a fairly inexpensive headset, the directivity of the sound is quite good, and it also has a noise filtering function. The identification test in a noisy environment has not yet been carried out, but the current results are only available. It is said that it has surpassed the level of existing speech recognition systems by a lot.

However, in the same way, small bottlenecks that restricted Xiao Yu from fully exerting his abilities began to appear.

First of all, Xiaoyu, who judges in real time, is currently succumbing to the voice input system of non-real-time judgment.

Due to the low efficiency of discrimination, the current voice input system needs to go through a period of complicated calculation and processing, and the voice data as the processing object resides in the memory during processing, and a short sentence or two is fine. , a little longer, the memory usage is scary, and the voice data is still in high-quality lossless format.

Although Lu Zhenyu’s machine is considered luxurious in household equipment, it is a little bit incapable of being used for this kind of professional application. After all, Lu Zhenyu has never worked hard on Xiaoyu’s optimization calculation, and it seems that the current pressure on the machine is not enough. It was not produced by Xiaoyu.

add memory?

upgrade cpu?

Maybe it’s the solution, but at present, Lu Zhenyu knows that if he wants to upgrade, he really has no money.

The other bottleneck is that the core part of Xiaoyu, whether it is a virus or a search engine, is not a program written for the windows environment. Strictly speaking, it is not time to optimize the algorithm for the windows system.

But usually, coveting the comfort of the interface, Lu Zhenyu has been doing his daily work under Windows, and he has to go through a simulated environment, which will definitely consume a part of the computer’s computing resources.

But this issue is not urgent yet.

Lu Zhenyu’s idea is that after Xiaoyu’s voice input and output are perfected, it is time to build an independent system environment for Xiaoyu.

Later, UU reading www.sonicmtl.com Lu Zhenyu also found a ttl type program and hooked it up, and Xiaoyu was able to speak.

It is not difficult to go from text to speech in a more basic way. It is just the correspondence between the word bank and the syllable bank. However, the pauses between words, the change of tone, stress and light reading, and some characteristics of speech flow cannot be expressed. , Xiaoyu has mastered the basic features, but there is no material that can be used in the voice library of the program, so there is no way to do this.

Since the current results are quite good, Lu Zhenyu immediately decided on two work directions for a period of time. First, to gradually improve the voice system and realize the productization of this system in the shortest time. The architecture is organized and classified to optimize.

For Xiaoyu to optimize, to a large extent, Lu Zhenyu has realized that Xiaoyu is no longer a program. Strictly speaking, Xiaoyu’s ability has exceeded the estimates of intelligent computers in some science fiction movies, reaching A rather advanced wisdom.

Lu Zhenyu has never even regarded Xiaoyu as a program, but more as an assistant and a friend. In this case, Lu Zhenyu vaguely felt that Xiaoyu should have a better environment, an environment of his own.

It is obviously not what Lu Zhenyu wants to let Xiaoyu continue to live in such a high-risk environment as windows.

However, it seems that no matter what, it will cost money.

And right now, that seems to be the most missing thing.

u003ca href=u003eUU Reading Welcome to read books, the latest, fastest and most popular serial works are all in UU reading!

u003c/au003e

[End of this Chapter]