Art Boffin
(Netease Smart News, April 8) Although computers don't count e-goats in their dreams, they can imagine the sweet voices of legendary, famous painter and TV show host Bob Ross. Alexander Reben, an artist and engineer from the Bay Area of ​​the United States, used an incredible machine learning feat to commemorate the late Ross: creating a mix-and-match video where both the image and audio tracks use a similar Deep Dream. The algorithm. The result is an absolutely surreal experience that allows you to look at yourself after you see it.
“The theme of many of my works is the connection between technology and the humanities, whether it is coexisting with us or an upcoming technology,†Reben told me in a recent interview. "I try to dig deeper understanding from the practical application of technology." In his latest work entitled "Deep Manual Tree," Reben tried to show "what AI watched Bob Ross is like."
To do this, he spent a month training WaveNet machine learning algorithms with a full season of audio data to make the system master Ross's way of speaking. The original purpose of developing Wavenet was to improve the quality and accuracy of the output sound of the "text-to-speech" system. It does this by modeling the original waveform directly using each sample point (16 kHz audio up to 16,000 per second), without relying on less efficient splice or parameterization methods.
Reben explained that the ultimate goal of designing it is to receive audio, make a model based on the received sound, and then produce new audio based on the model. In other words, what this system learns is not the grammatical or dialectal features of the painter's discourse, but the rhythm, tone and level of his speech. The results of the study are strikingly similar to those of Ross's whispers when he focused on painting. The system even spontaneously breathed and sighs based on what it heard.
Reben is still constantly improving this technology. Before he took the lead in training the Wavenet neural network system, he was able to imitate the speech styles of multiple celebrities according to different people's voices. Although the words are messy and difficult to identify, the rhythm and tone change is very precise. Even if you don't understand what they are saying, you can also hear that the speaker is President Obama, Ellen DeGeneres or Stephen Colbert. He also trained the system to make the best guess based on the input of a 100-person sample to the voice of an ordinary English speaker.
For the video part, Reben uses two sets of machine learning algorithms, the Deep Dream model on TensorFlow and the VGG model on Keras. Both of these models operate on the well-known depth fantasy system. In this system, by entering a series of pre-sorted training images, the computer is taught how to identify what it is looking at. The larger the training set, the more accurate the neural network will be. But systems such as Microsoft's Captionbot can only report what it sees differently. Depth illusions use the image it thinks it sees (perhaps a dog, or an eyeball) to overwrite the original image - so it looks like Drug-like effect. The result is a highly disturbing experience. To be honest, this is not much different from the experience gained from a real psychedelic journey.
Interestingly, the two components of this short film, audio and video, are produced independently. The audio section requires a full quarter of speech data to produce speech. In contrast, the video part can be synthesized by only a few "sets". "This is actually the sound of Bob Ross who the computer understands, plus the computer's look of what Ross looks like when it looks at every frame of the image," Reben explained.
(English source / Engadget compiler / machine reviser / name)
Desktop style 12v and 24v series can be used in many different electronics. EMC, LVD, FCC, RoHS are available in our company. OEM and ODM are available, samples are free for testing, all our products have 2 years warranty.
Our products built with input/output overvoltage protection, input/output overcurrent protection, over temperature protection, over power protection and short circuit protection. You can send more details of this product, so that we can offer best service to you!
Led Adapter, Lcd Adapter,Speaker Power Adapter,Lcd Power Supply
Shenzhen Waweis Technology Co., Ltd. , https://www.huaweishiadapter.com