IFLYTEK has won the international spoken machine t

  • Detail

After winning eight "world firsts" this year, iFLYTEK made another effort. In the just concluded 2018 international spoken machine translation evaluation competition, iFLYTEK defeated jd.com, Alibaba, Sogou and other strong rivals to win the championship in the end-to-end model competition of English German speech translation tasks

the project participated by USTC nel in this competition: SLT (spoke language translation): baseline and end-to-end

what kind of experience is it to turn the ability to win the honor of "world champion" into daily life? IFLYTEK's good performance in the international top oral machine translation evaluation competition is not only a major breakthrough it has made in the competition since it won the championship in Chinese English translation in 2014, but also the Ninth "world champion" the company has won this year, which fully proves its technical hard power in the field of AI and its international leading position in the field of voice translation

from 2004 to 2018, iwslt has held 15 competitions, each of which is a grand event. Each competition attracted well-known enterprises and research institutions in the field of machine translation from all over the world, including domestic companies such as iFLYTEK and Alibaba, Irish adapt center, Johns Hopkins University in the United States, Karlsruhe Institute of technology in Germany, University of Helsinki in Finland, University of Edinburgh in the United Kingdom, apptek in the United States and many other internationally renowned universities, research institutions and companies

what is the "gold content" of this "champion" on the new track in the field of AI translation

it is reported that the oral machine translation task of this iwslt 2018 mainly evaluates the speech translation ability from English to German based on the specified data set, which is divided into two tracks: baseline model track and end-to-end track

if Sogou, which won the first place in the baseline model track evaluation, is praised, iFLYTEK, which won the championship in the end-to-end track evaluation, deserves more attention. Because as a newly added track in this evaluation, this is a technological innovation realized through the core source technological innovation

because the traditional baseline model scheme is to recognize the input speech first to get the text results, and then carry out text machine translation. In practical application, due to the existence of speech recognition errors in complex environments, error diffusion affects the final speech translation performance; The end-to-end modeling scheme is based on the powerful modeling ability of neural network, and the input speech directly outputs the target translation

to this extent, iFLYTEK won the second place in the traditional baseline model, but it was even more surprising to win the first place in the latest end-to-end model. On this new track, the "champion" has a heavier "weight" and is more eye-catching

the tide is the trend. In the past 19 years, iFLYTEK has continued to lead the industry with technological innovation.

iFLYTEK, which has been deeply involved in the field of intelligent speech and artificial intelligence for 19 years, has always adhered to the source technological innovation based on the profound accumulation of speech recognition and machine translation, and has successfully applied the latest in-depth learning and research results to stop overloaded speech recognition tasks and various products, Continue to land in product iteration and upgrading

this has been vividly demonstrated in iFLYTEK's products. The official designated translator of Boao Forum for Asia - iFLYTEK translator 2.0, which organizes the annual special energy-saving supervision and special supervision of major national industries, is equipped with a number of world-leading core technologies such as speech recognition, speech synthesis, image recognition, machine understanding and four arrays to ensure "clear", "understandable", "accurate translation" and "beautiful pronunciation", and defines the four major standards of AI translation

more importantly, iFLYTEK translator 2.0 has more powerful functions than this: it realizes the real-time translation between Chinese and 33 languages around the world, has a brand-new inmt translation engine, the comprehensibility of Chinese English translation reaches the level of university level 6, as well as the personalized function of offline translation, photo translation, global top-ranking rich inspection of whether the friction parts should be filled with lubricating oil, covering the needs of the whole scene

what is worth mentioning is its exclusive dialect recognition ability. At present, it supports the real-time translation of Cantonese, Sichuan dialect, northeast dialect, Henan dialect to English, and will continue to launch more dialects in the future

in artificial intelligence, the "weathervane" of future technology, iFLYTEK can be said to firmly grasp the "tuyere" and continue to practice technological innovation and breakthroughs, only to provide consumers with the best experience. The technical breakthrough in the task of the iwslt2018 speech translation competition will also be combined with iFLYTEK's core AI technology to further improve the speech translation performance of subsequent iterative products

in the great historical process of artificial intelligence, the continuous pursuit of technological innovation is the surging direction. In the future, let's also look forward to iFLYTEK how to light up the whole industry with the light of AI

Disclaimer: the above content is transferred from other media by our website, and the relevant information is only for the purpose of transmitting more information, which does not represent our view, nor does it mean that our website agrees with its view or confirms the authenticity of its content

after Sohu technology won the eight "world firsts" this year, iFLYTEK made another effort. In the just concluded 2018 international spoken machine translation evaluation competition, iFLYTEK's result was that a new thermosetting material flew in the end-to-end model competition of English German speech translation tasks, defeating JD and a

Copyright © 2011 JIN SHI