Chapter 69: Voice temperature

浩 Wu Hao shook his head with a smile and said, "No, it is just an unfinished product. There are still many problems that need to be solved.
For example, in the dialogue just now, it is more difficult to understand and deal with vague contexts. "
"Vague context?"
Zou Xiaodong stunned for a moment, and soon came to understand: "This seems to be difficult for us to understand, let alone the machine program.
Boss, I don't quite understand. Speech recognition and speech dialogue are currently being done by most technology companies, and the results are good.
The recognition degree of these speech software for our normal speech is also very high, which can basically reach more than 99%.
But the response speed of these softwares is far less than that of our technology. The ability to understand is not as strong as it is, and the processing power of Lenovo is not comparable.
In addition, in terms of voice dialogue, how do you do that, so that the language of the machine can be so close to the voice of a real person.
We need to know that human hearing is still very sensitive, and whether it is human or machine program sound can be distinguished quickly. "
Wu Hao heard a lot of questions from Zou Xiaodong and asked him, "What do you think is the biggest difference between a real voice and an AI voice?"
Zou Xiaodong thought for a moment, then replied, "Is there a lack of peace?"
Wu Hao shook his head and said, "This is not the most critical. In fact, some voice software on the market can already simply calm down."
"That is……"
Wu Hao looked at Zou Xiaodong's inexplicable look and said with a smile: "Emotions, all the voice program software on the market now have no emotions."
感情 "Emotions, what's this joke, how can the program have emotions, this is a talent." Zou Xiaodong shook his head and could not understand.
Wu Hao smiled, and then controlled the computer to display a structural diagram on the big screen and said, "It's not the language but the emotion.
When we are speaking, the other party can clearly perceive the emotional changes when we speak. This is the emotion and this is also the language temperature.
What's more, the language program reacts according to a fixed formula. So it can't understand the temperature of each sentence, naturally there is no temperature in generating speech.
What we need to do is to add an understanding of the language vocabulary environment in the process of speech recognition stereotypes, and analyze the temperature of the discourse and the speaker's emotional changes from different tones. "
"I still can't understand how people's emotions change when they speak and how the program can capture them. You need to know that sometimes slight changes in language and tone can show two very different meanings and two emotions. How the machine can tell. "Zou Xiaodong said his doubts.
Wu Hao smiled while demonstrating the content on the screen, and replied to him: "This is the use of AI technology. Everyone's language is different, and the expression of emotions is also ever-changing. If we use the traditional method, we need to change these Grab, collect, analyze, and define the language intonation context. If this is the case, the workload can be too much.
So the learning and evolution ability of AI technology allowed me to find ideas. We can train a set of basic AI voice programs by capturing the loving voice information on the Internet.
Of course, this is just a sample of the basic program, we need to make corresponding adjustments according to the habits of users. Let the program learn to adapt to the user, the longer the user uses it, the more accurate the recognition and understanding of the AI ​​recognition program. "
Speaking of this, Wu Hao laughed: "This is actually very similar to the way we live in real society. After two strangers get along, both sides gradually figure out how to adapt to each other.
The longer the time, the more familiar the two parties will be. Even one party can receive and understand a simple word, gesture or look accurately. This is called tacit understanding.
What we need to do is to cultivate the tacit understanding of procedures and people, but the user is difficult to change, and can only have a subtle influence. So we have to start with the program software, let it adapt to the user, and change the user implicitly.
Only in this way, human-computer interaction will be more tacit.
This is also the reason why I couldn't understand my ambiguous context when I was talking with 10. It didn't adapt to my speaking habits, so it didn't understand what the ambiguous words I said meant.
什么 What, how many, how much, so, where, random, these uncertain and ambiguous words, the program is difficult to understand and deal with. And this requires us to give a basic definition of these words. This definition cannot be rigid and rigid, but also has to be modified according to the context of the user. "
After saying this, Wu Hao looked at Zou Xiaodong and said positively: "Only after the program understands the emotional temperature in our real words can the program simulate a voice similar to a real person's speech."
"Anyway, this is a major breakthrough in the field of AI voice technology. I think this technology will definitely shake the world once it is released, but it represents the real arrival of this intelligent voice era ~ EbookFREE.me ~ To be honest, I can't wait any longer. "Zou Xiaodong licked his dry lips and said excitedly.
浩 Wu Hao waved his hands and said, "It's not as exaggerated as you said, but it is indeed a major breakthrough in technology."
"Boss, do you intend to use this technology directly for the mass consumer market, or do you cooperate with corporate users, sell technology and related patents, or provide services for them with open source relaxation?" Zou Xiaodong wondered at him. This is a heavyweight technology, no matter who you work with, it will bring a huge shock to the industry.
"What do you think?" Wu Hao did not answer directly, but asked back.
Zou Xiaodong thought about it, and then said to Wu Hao seriously: "A company that wants to become bigger and stronger cannot be confined to a single field. Cooperation with enterprises can save a lot of things, but the risks are great. With more advanced technology, we face the risk of being abandoned.
So I think we should develop the mass market, use this technology to build our brand among the people, and expand our influence. Only in this way can we reduce unnecessary troubles and resistance in future development. "
"The analysis is in place, but the market has huge potential. Monopoly alone is definitely not enough. We still need to cooperate with those companies. Of course, we cannot lag behind in the mass market.
So I'm going to do both, and this smart voice assistant is made for the mass market. How about, put out the video I just showed, what do you think will be the reaction in society and industry. Wu Hao asked with a smile.
"You mean ... haha, I look forward to it!"