Google's new AI can voice like a human now
If you are a fan of science fiction movies such as "The Terminator" or "I, Robot", you will probably realize that the technologies in these films are gradually being realized at the present time. The latest researches from Google show that the technological prospects that you think can only occur in movies are getting more closer to reality than ever before.
In particular, the research titled "Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions" has unveiled a new Google text-to-speech system called Tacotron 2. It is known that this system has the ability to counterfeit human voice.
In order to achieve this succeed, Tacotron 2 uses a pair of neural networks with different roles: a network produces visual images of specific audio frequencies and the other network (WaveNet) reproduces the visual data as audio.
Google’s Tacotron 2 text-to-speech tech sounds like a human voice
In addition, Google has launched a website to further illustrate what the technology can do in practice. They have provided many examples of how Tacotron 2 handles homophones, such as distinguishing the meaning of the word "present" between the noun form and the verb form. Besides, this system also solves the problems of intonation as well as many difficult words that even humans can rarely use.
In the last part of the article, Google also demonstrated the superiority of Tacotron 2 by offering multiple pairs of identical audio files created by humans and their new AI system, which is difficult to distinguish. You can experience and evaluate Google's new AI technology here.
On the other hand, this technology is a small part of the larger mission that Google is pursuing: enhancing Google Assistant's dialogue capabilities. This virtual assistant is the key AI factor behind the Google Home - Google's new potential business. And the Tacotron 2 system really fits that line of equipment.
Google is developing Google Assistant to be more “human”
At this time, Google Assistant is definitely more complete and more efficient than ever before. However, Google's report shows that the virtual assistant is likely to be more human-like in the near future.
In fact, the distance between AI and humans is still huge, and it is hard for a machine to communicate like humans. The elements of nuance, context, emotions change constantly during the conversation and play a very important role. That is the biggest obstacle that AI needs to overcome to have more "personalities".
However, with the current pace of development, AI will be more advanced and will be able to talk naturally in the near future.
By: Frank Richardson