Qianjiang Evening News·Hourly News Correspondent Wang Huihua Correspondent Zhang Haifeng
According to Beiqing.com, my country will develop new industrial chain layouts in frontier fields such as artificial intelligence.
At the beginning of school, the reporter also received questions from some students: Are some of the artificial intelligence products that are being used for learning English really smart teaching products? Are the siri in the Apple mobile phone and the Tmall genie in many homes the same as the technology in the intelligent foreign language teaching products... Industry professor Dr. Qifeng Zhu. Dr. Zhu studied at Tsinghua University , Chinese Academy of Sciences, University of California, Los Angeles , Berkeley. He is also an expert in intelligent teaching and has served on three national information education research projects. He independently developed the world's leading artificial intelligence voice technology and founded Qingrui Education. The most eye-catching products-" oral 100 smart learning space" and "intelligent listening and speaking coach" are widely used throughout the country. This time he brings you answers related to artificial intelligence.
human-machine dialogue, one of the standard configurations of artificial intelligence
In the past 20 years, the information age has surpassed the digital age, cloud technology age, and big data age, and has come to the age of artificial intelligence.
We can actually understand from the perspective of learning more connected with classmates. Every day, students can use artificial intelligence-related products, and even in daily learning, there will be many experiences, such as speech recognition, speech analysis, speech synthesis, oral assessment, image recognition, handwriting recognition, natural language understanding, Human-computer dialogue, composition correction, etc., and the level of new technology is constantly improving.
And these teaching products that we come into contact with every day have also experienced the " e-school bag " and exercise CDs, network multimedia learning content, precise analysis suggestions through big data analysis, and teaching content push,Until now, the stage of personalized interactive training of artificial intelligence. For example, after R&D personnel have applied a variety of artificial intelligence technologies to foreign language teaching, they have also brought students a new and effective form of training: anthropomorphic intelligent interactive language ability training.
The first generation of information-based teaching products are generally traditional teaching content electronic, and intelligent interactive training is completely personalized, according to what students say, intelligent teaching products can be like real people, can listen, speak, Can see, talk, chat, evaluate and correct, and improve students' language skills through personalized intelligent interaction.
However, students must be more concerned about how this new type of artificial intelligence interactive language training can help them in foreign language learning and other aspects?
After reading the questions and hopes of the students, we also found that one of the biggest characteristics of artificial intelligence equipment is that it can communicate with students, which is what we often call "human-computer dialogue."
smart devices, how do they learn to talk
So how does artificial intelligence work? For more than 20 years, the working method of artificial intelligence has evolved from expert systems and decision trees to statistical models and the latest multi-layer neural network models-also known as "deep learning." This deep learning is completely different from what we think of as deep learning. It refers to a big data model method in which the number of layers of artificial neural networks that simulate the connection of brain neurons may be as many as a dozen or even hundreds.
For example, in the application of human-machine intelligent dialogue, speech recognition technology must be used first. In order for the machine (smart teacher) to understand human speech, R&D personnel need to collect tens of thousands or even hundreds of thousands of hours of student speech to "train" this multilayer neural network model.
How much storage does 100,000 hours of student voice need? One minute of sound requires about one megabyte of storage, and 100,000 hours of sound requires 6TB (6000G) of storage space. The entire training process requires multiple computers and multiple CPUs and GPUs to be used together, and distributed collaborative work runs continuously for about a week. It can be seen that machine learning is a very computationally intensive task.
In addition to using speech recognition technology,Smart machines have to be trained in pronunciation analysis techniques.
In this task, the machine already knows what the student said, and then it needs to judge whether the student's speech is similar to the standard voice. The goal of speech recognition task is to tolerate and understand all kinds of bad pronunciation, and pronunciation analysis is to pick out the problems in pronunciation. This seemingly simple task is actually not simple-one can easily judge the A said by a girl and the A said by an adult male as "the same pronunciation", and cleverly ignore that they are actually caused by differences in gender and age. The difference is huge, but it is a challenge for machine intelligence.
In various intelligent interactive training tasks, speech synthesis technology also plays an important role. With this technology, computers can speak to people, not only to prevent students from always reading the text on the screen, but also to interact more naturally and protect their eyesight. Different technical solutions can be used behind the speech synthesis technology, including waveform splicing technology: directly record a person's voice read aloud in the recording studio, and store it as a tens or hundreds of megabytes of sound waveform warehouse containing different contexts. Then you can search for the correct context waveforms for splicing according to the needs of the synthesis task; you can also use the "vocal tract parameter model" to simulate the process of airflow through the vocal cord vibration and the vocal organs to produce the final speech waveform inside the computer. This method does not need to store the sound directly, but it needs to analyze the vocal tract model of the person's articulator from the sound.
Questions
In many science fiction movies, students must have seen the ubiquitous artificial intelligence in the future world, and it is smarter than humans can imagine now, and even produces autonomous consciousness and emotion. But will all this happen?
In fact, artificial intelligence is actually not that smart, but it has achieved "diligence to make up for the weakness": through the enhancement of computing power and diligent big data learning, computers can catch up and even surpass humans.
However, to give a simple example: a human child can learn what a cat is by seeing a few cats, and a computer may have to look at hundreds of cats to learn what a cat is. It can be seen that the human learning method itself still has its own uniqueness and efficiency.
Today, mankind has just stepped into the threshold of the era of artificial intelligence.But it is foreseeable that artificial intelligence technology will eventually serve mankind. Nowadays, the production of excellent intelligent teaching products requires the use of multiple artificial intelligence technologies, and each technology can adopt different technical solutions according to different application goals. Therefore, the development and research and development of artificial intelligence technology in the future will require a large number of talents. I wonder if the students are also willing to join in in the future?
This time, Dr. Qifeng Zhu also asked two open-ended thinking questions for the students. After scanning the QR code on the left, you are welcome to leave a message in the article message area for interaction, and mark your name, contact information and express delivery. Piece address. We will select 100 excellent commenters in the message area and give exquisite small gifts-
think about it
1. What do you think of the importance of interactive language training for foreign language learning?
2. What do you think is the biggest advantage of intelligent interactive language training? What is the biggest disadvantage?
This article is the original work of Qianjiang Evening News. Without permission, it is forbidden to reprint, copy, extract, rewrite, and spread the copyright of all works, otherwise, the newspaper will pursue the legal responsibility of the infringer through judicial channels.
Source: Qianjiang Evening News·Hourly News
.
