This month, the voice of South Korean superstar Kim Kwang-seok will be returning to national television.
The performance, which will feature all-new material, is literally death-defying as the iconic folk singer passed away 25 years ago in 1996. It would not have been possible without the use of artificial intelligence (AI) technology.
This isn’t the first time South Korea has resurrected the voices of late singers— holograms of popular singers have appeared in New Year concerts, and have even performed virtually alongside living artists.
How was AI used to recreate Kim Kwang-seok’s voice? Let’s take a look.
The Science Behind the Voice
We have all heard the digital voices of Amazon's Alexa or Apple's Siri, or voice assistants in the cars giving us directions to our destination.
These applications of AI in our everyday lives have taken decades to perfect. Machines were trained to mimic human voices, fed a lot of data, and were taught to self-correct if they gave out incorrect answers.
Natural Language Processing (NLP for short) is the machine's ability to read content created by humans. Natural Language Generation (or NLG) is the ability of machines to create original content that can be understood by humans. Both NLP and NLG are a part of AI technology. And over time, machines like Alexa have been taught the ability to hold a conversation with a human!
To recreate Kim's iconic voice, the AI program listened to 20 of his songs after which it was able to replicate the singer’s voice with near-perfect accuracy and feeling. AI Kim will not only be singing his originals but will also perform songs by other artists. In fact, his hologram will be performing “I Miss You”, a 2002 ballad by another K-pop icon who passed away six years before.
Viewers might be excited to see their favorite past singers come alive. But the increasing influence of AI on the music industry has raised concerns over originality and ownership. Who is the legal owner? The AI, or its programmer?
Even more alarming is AI technology’s use for far more malicious purposes. Deepfake videos utilize AI technology called deep learning to generate convincing, but fake graphics of unreal events. Deepfakes can be entertaining, such as the 2019 video of Game of Thrones character Jon Snow apologizing for the series’ disappointing finale.
But they shouldn’t be mistaken as completely harmless either. In March 2019, a top executive in a German company was fooled into transferring £200,000 to a Hungarian bank account by scammers who used audio software to simulate the CEO’s voice. In an era when it’s become increasingly difficult to separate what’s real and what isn’t, deepfakes complicate the situation.
Countries and international organizations are in the process of creating guidelines for ethical and intellectual property standards in AI. What do you think? How can we enjoy AI and still keep it safe and ethical?
Sources: Reuters, CNN, Guardian, Britannica