Korean Researchers Develop Deep - Learning technology for telecommunications

인공지능으로 끊어진 통화음도 되살린다 Artificial intelligence or AI. Able to learn to find answers on its own, the technology is being applied in various fields. Park Se-young takes a look at one example in Korea: preventing phone calls from being garbled when the signal is weak. Back in 2012, computers taught themselves to recognize a cat based on an analysis of ten million images. Now, the same approach is being applied to speech recognition technology for telecommunications... by a team of researchers at Hanyang University. They've developed a program for recovering parts of disconnected phone calls... but that's just the start. "With deep learning, we'll be able to create humanoid robots with advanced speech recognition." In a recent demonstration of the software, a cell signal keeps cutting in and out during a phone call. "...ate ...handful ...and fruit" At first, the sentence is hard to understand. But then, the researchers tried the phone call again using their program. "I ate a handful of snacks and fruit." The sentence sounds much smoother and is more understandable. It was recovered using deep learning, which finds patterns from a massive amount of data and estimates the missing parts. In a comparison of the audio graphs from an undamaged file and the recovered file,... there's almost no difference between the two. Researchers are still in the initial phase of development, but they're aiming to achieve better speech quality and recognition down the road. Their study was published in February in the journal IEEE Transactions. Park Se-young, Arirang News. Visit ‘Arirang News’ Official Pages Facebook(NEWS): ****** Homepage: ****** Facebook: ****** Twitter: ****** Instagram: ******

Artificial intelligence or AI. Able to learn to find answers on its own, the technology is being applied in various fields. Park Se-young takes a look at one example in Korea: preventing phone calls from being garbled when the signal is weak. Back in 2012, computers taught themselves to recognize a cat based on an analysis of ten million images. Now, the same approach is being applied to speech recognition technology for telecommunications... by a team of researchers at Hanyang University. They've developed a program for recovering parts of disconnected phone calls... but that's just the start. "With deep learning, we'll be able to create humanoid robots with advanced speech recognition." In a recent demonstration of the software, a cell signal keeps cutting in and out during a phone call. "...ate ...handful ...and fruit" At first, the sentence is hard to understand. But then, the researchers tried the phone call again using their program. "I ate a handful of snacks and fruit." The sentence sounds much smoother and is more understandable. It was recovered using deep learning, which finds patterns from a massive amount of data and estimates the missing parts. In a comparison of the audio graphs from an undamaged file and the recovered file,... there's almost no difference between the two. Researchers are still in the initial phase of development, but they're aiming to achieve better speech quality and recognition down the road. Their study was published in February in the journal IEEE Transactions. Park Se-young, Arirang News.