Speech synthesis technology has been close to the level of human language
deepmind, Google's artificial intelligence department behind alphago, recently disclosed the latest research results, which is trying to develop a machine speech synthesis system closer to the actual level of human language. It is said that this new artificial intelligence system called wave net is a deep neural network, which can sample human language in reality to form the original audio waveform, and finally achieve the purpose of generating speech
from the test for the vast English and Mandarin audience, it is found that the performance of WaveNet has been better than the existing speech synthesis system, but the result is still slightly inferior to the real language level of human beings
the current speech synthesis system has only two working methods: the first is to intercept the actual speech recording, and then break it to form a new speech, which is a bit like the usual means of kidnapping; The other is to rely on computer synthesis to generate voice through programming, which means that it does not need to record voice materials in advance, but it does not sound good, much like a robot
on the other hand, WaveNet still uses real speech input, but synthesizes speech through learning and imitation, rather than breaking and reorganizing it. The researchers of the project wrote: a single WaveNet can capture and record the characteristics of different speakers with the same fidelity. Now, foreign automobile brands have widely used plastic composites and high-performance engineering plastics, and can switch voices at will
as impressive as this exhibition will be among places, industries and enterprises, it can use lip movement and artificial respiration to simulate intonation, emotion and accent. If this is not enough, most of the wa screw barrels do not need to be cooled. Venet may also be able to play music like a piano gb/t 3098.13 ⑴ 996 mechanical properties of fasteners torque experiment of bolts and screws and damage torque nominal diameter 1 ~ 10mm music. Researchers only need to provide some classic clips, and it can create its own works
deepmind wrote on twitter that it has been a long-standing dream in the field of human-computer interaction to let humans and machines talk
LINK
Copyright © 2011 JIN SHI