You are a black technology boss, you didn't say it sooner!

Xie Jin didn't know how Lou Qingyan discovered his thoughts, but, to be honest, he didn't quite believe what the other party said.

Maybe it was just to comfort him.

Of course he doesn't know that the earth really doesn't need algorithm innovation, because the existing neural network model is actually the first generation technology of strong AI.

The first strong artificial intelligence that appeared in the interstellar world was formed by stacking layers of neural networks.It is essentially a learning machine that generalizes human logic from many sample events and writes it into a database or algorithm.

When the magnitude of the system exceeds a certain critical value, the entire logic volume will be extremely large, covering all human thinking directions, and the behavior mode of the machine will be infinitely close to human beings.From this, it can be considered that quantitative changes lead to qualitative changes, and machines gain the ability to think independently.

To achieve this goal, the artificially constructed machine neural network must have a volume no less than that of the human brain.But how huge is the number of brain neurons?Even with the last kilobyte of capacity squeezed out of a supercomputer, this structure cannot be simulated.

The earth needs hardware updates, and photonic computers barely meet the requirements, and the ideal is a quantum computer.However, quantum computers are a bubble bigger than strong AI. Its development requires at least three prerequisites: controllable nuclear fusion, brain domain development technology, and basic scientific breakthroughs.

As for why, it was too troublesome to explain, and Lou Qingyan didn't intend to tell Xie Jin.

Xie Jin didn't struggle for too long, and quickly brought him to a workstation.

"Chairman, since you want to watch the electronic soundtrack project, why not take a look at the demo first. This version is already very mature, as long as some optimization work is done, it can be put into market operation." He clicked on the mouse.

A song played on the stereo, the girl's voice was very sweet, the pitch was a bit weird, and the connection was mechanical, it sounded like Siri was singing.

However, compared to the vocaloid series, it is already very natural.

Vocaliod, the famous voice synthesis software, is the basis for the birth of the super popular idol Hatsune Miku.

It is said that it was originally recorded as a demo for the music industry, but after its release, there were very few people interested in it.The producer slapped his head and opened his mind: "We don't make music software, but virtual singers." Thus, Hatsune Miku was born, and the concept of virtual idols came into people's sight.

Xie Jin said: "When designing the software framework, we compared several competing products, among which vocaliod and utau are the most popular. Their implementation principles are similar, pre-recording human voices, arranging and combining human voices, and adding tones to form A song."

"We picked up the framework of the two softwares, and finally decided to abandon this purely mechanical synthesis and change it to a bionic one. That is, AI software based on machine learning algorithms. We named it electronic vocal cords."

"Because there is no precedent, the research and development period is very difficult, but after the first version came out, it gave people quite a surprise. The finished song, at worst, can reach the level of a singer." He clicked the mouse, facing the building. Qing Yan said, "Chairman, let's listen to the first sentence it said."

A clear male voice came from the speaker.

"hello, world."

Lou Qingyan suddenly realized: "Is this my voice?"

"It was decided by everyone's vote." Xie Jin looked over teasingly.

For some reason, after being distracted, he felt that Lou Qingyan was a little kind.

"It's okay, it's quite natural."

"This is the credit of the AI ​​algorithm." Xie Jin said, "We have designed a deep learning algorithm. The software will extract the features of the sound source library, use the extracted feature information for training, and finally generate a unique sound source library for each group. Data model. Based on this model, the synthesized speech will be much more natural."

"This is not the most ideal. The most ideal kind of synthetic voice does not need to combine human voices, but relies on the high intelligence of AI algorithms to generate voices out of thin air. But..."

"But you don't have this technology." Lou Qingyan said.

He thought for a while and described: "The essence of sound is vibration. In the stereo, it is the vibration of the diaphragm. This vibration is caused by the electromagnetic field, and the control method of the electromagnetic field is whether the coil is energized or not. It is 1 when it is energized, and it is energized when it is not energized." 0."

"So, the essence of computer sound is a string of binary numbers, 1 and 0 in different combinations. This is called a digital signal."

"The most ideal speech synthesis is not combining syllables, but directly combining digital signals."

Lou Qingyan: "Am I right?"

Xie Jin was really surprised.

He always thought that Lou Qingyan was an ignorant dude, the kind that he couldn't understand even if everyone understood the principles of computers.

I didn't expect him to really understand.

Before Lou Qingyan finished speaking, he continued: "Theoretically, AI software can grasp the laws of digital signals through deep learning, build a data model, and then generate voice based on the model."

"However, this requires a major premise-machine learning algorithms can perform feature extraction directly from the digital layer."

"Can you?" he asked.

Xie Jin was convinced, why does this person always ask the most important points all at once.

"There is really no such technology now. The feature extraction algorithm we designed is mainly aimed at the waveform performance, frequency, and phase of the sound." He said simply.

Lou Qingyan added: "It's not just the lack of this technology. After feature extraction, do you know how to summarize? Do you know how to train the model? Can you write algorithms? Have you imagined how huge the amount of calculation is?"

Xie Jin: "..."

Lou Qingyan realized that he had said a little bit, so he patted him on the shoulder and encouraged him: "But you have done a good job, and it is amazing to reach this level... Well, what is the next step?"

Xie Jin: "..." It can be seen that you are being perfunctory.

"The structure of the software is a bit bloated, and it can only run smoothly on a workstation. The next step is to consider optimization."

"Okay." Lou Qingyan asked casually.

He doesn't care about the software itself, and knowing that there is already a prototype of AI speech synthesis, his purpose of coming here has been achieved.

The electronic soundtrack developed by Xie Jin is actually recording human voices in advance, and then arrange and combine them.It's just that the data model based on AI algorithm training will modify the audio waveform to make the sound smooth, natural and more real.

The limitation of this software is the algorithm.There is no precedent for feature extraction in the audio field. Xie Jin’s team must thoroughly study digital audio, and then design new algorithms based on its characteristics—not only the logical framework, but also the evil mathematical model.This is not something that can be done overnight.

But as for Lou Qingyan, he has something ready in his mind.

In the interstellar era, sound synthesis is a very popular technology. Cosmic people have long been able to generate various sounds at the coding level, and a new profession called audio writer has emerged.

Analogous to literary writers, audio writers are the ones who create radio plays directly.

The sound synthesis algorithm is an enlightenment case for beginners to learn omnimechanical programming.You can see how simple it is.It can also be realized on the earth, the condition is several sets of new mathematical models and a large computer.

Jiahui has a large computer, which is used as an app server. It seems that it is no big deal to temporarily shut down the server to calculate the software.

Lou Qingyan has only one person, so it is impossible to write a complete software, but he has Xie Jin's code foundation, as long as he implements those mathematical models and uses a large computer to quickly complete machine learning calculations, he can achieve his goal.

Well, he just wanted Lou Haochen to "say" a few words.

Xie Jin didn't know what kind of dark criminal process his boss was planning. He looked at Lou Qingyan, and his heart had completely changed towards him.

The dragon is trapped in the shoal, and the tiger falls in Pingyang, but it seems calm and composed.When he is down and down, he does not show any embarrassment, and his posture is still so tough in the face of the negative ridicule of employees.He understands computers and intelligent technology, and he is not the kind of ignorant waste that everyone thinks.Most importantly, he actually understood his thoughts.

Xie Jin strongly hoped that Jiahui could get through this difficult time.Because it was the first time he realized with real emotion that Lou Qingyan was a good boss.

He secretly glanced at the other party, just when he heard him say: "That's it, the tour is over. The next step is to learn C language in a day."

Xie Jin: "...?????"