Week 3


( 1 ) Week 3 Assignment :

Find an example of synthetic media (marketing, art, entertainment, something else) Go beyond deep fakes, find novel examples if possible. Write a blog post about the example(s) you found and answer the following questions:

Why do you consider this synthetic media?
What can you find out about how this was made (if you can)?
If it’s machine learning-based can you find information about the data set used to train it?
What are the ethical ramifications of this specific example (if any)?

On this topic, I want to talk about the use of ai in the text part. The first thing that came to mind was the Grammarly, which I usually use. It uses AI to detect the grammatical correctness of the article and provide some suggestions on the sequence of sentences and word selection. Similar to this product is Wordtone, which is also enjoyable. He can select a whole paragraph and give you several different tones, and sometimes even translate it into sentences that are not the original meaning. These products sometimes make me feel that the computer knows better how common sentence patterns are formed, and I am curious about how language analysis is trained.

In addition to these two products, I have previously read articles about "Google Translate" on the Internet. In 2019, the Google team shared the Google Translatotron. In addition to translating, they are also trying to preserve the emotional message in the voice. The system will generate a sound map based on the received speech, and then the trained AI will analyze the map and directly generate a voice map of the target language according to the analysis results. And finally, the system will play the sound map with a "tone." In a few examples, I was amazed at the tone of voice that AI can simulate.

To answer the above questions, I think these products use AI in text and trained AI to create new text content, similar to the well-known deepfake, but in different fields. Secondly, r egarding the text machine learning part, I searched for information about Generative Pre-trained Transformer and Transformer. But I didn't read the announcements about which databases they said they were using when using these products. In terms of ethics, I think there are similar problems with deepfake, such as fake news, fake news, etc. Also, because Google Translatotron can have emotional voice translation, it faces the same moral problem of fabricating false voices (and emotions) as other AI voices.

listen to more audio samples: 

Introducing Translatotron: An End-to-End Speech-to-Speech Translation Model https://ai.googleblog.com/2019/05/introducing-translatotron-end-to-end.html

Week 3


( 2 )  Additional (Optional) Readings: