Sensitive ears and smart brain, see Sogou "friend" engine how to make the interaction more natural

Ears how a bosom friend

wangguanxiong· 2016-08-05 19:26:03

with the concept of artificial intelligence hot, intelligent voice began to capture a variety of intelligent terminals: voice input, voice navigation, voice and pictures voice dialing...... , however, the industry situation seems to be booming, many users have said that there is no real sense of the convenience brought by artificial intelligence. Is the machinery operating mode and response results are often disappointing, give an irrelevant answer regular dialogue process, also make people feel embarrassed and lose patience.


in order to improve the interaction because of not natural embarrassing experience, let the expression of information and access to more easily, recently, the artificial intelligence technology leader, announced the launch of Sogou voice interactive engine "friend". It is understood that the "salon" focus on natural interaction, not only has the sensitive ears, but also has a "smart brain". Style= color: "RGB (immediately following the title of the blue word attention can get more inside, a daily depth!)




[years of accumulation of data security, "bosom friend" recognition of the rapid progress of]


we know, voice interaction is divided into three ring section, the first is the second speech recognition, semantic understanding, speech synthesis is third. Speech recognition as the first step in speech interaction is the basis of understanding, thinking and interaction. Thus, it can be seen that a sensitive ear is very important to natural interaction. />



is worth mentioning is that the speech recognition engine uses a friend "the industry's most advanced" end-to-end "model, the technology can better solve the problems in the slur talk fast in the process of the user, at the same time in the training method used a kind of" active learning "method. Through the "end to end" model and hardware acceleration, Sogou speech recognition speed increased 3 times, recognition error rate is reduced by 30%, greatly reduced due to the accent and noisy environment caused by the error rate.


[the real solution to the user pain point," bosom friend "make interaction more natural]" color: style=


the voice interactive products there exist two problems: frequent error identification well, the interactive ability is insufficient, the result of user needs long time is in the interaction process of staring at the screen or manual operation too much, not real liberation of eyes and hands, let the interaction experience greatly reduced.


in order to solve these problems, Sogou "bosom friend" engine not only can support users to modify the error identification results in speech interaction with the user, also conducted several rounds of dialogue, a more accurate grasp of user needs.


first, relying on the Sogou input user dictionary library and the method of massive demolition, "friend" can support users to use natural language to correct errors in speech interaction process, and support the rich and the description method of substitution, insertion and deletion of hundreds of modified wrong grammar. For example, the user can say: "the 'Zhang' is amended as an early 'chapter', or the 'Zhang' is modified to the article 'chapter'"". At present, "bosom friend" changes to the success rate reached 80%, which is equivalent to a speech recognition error rate is further reduced by 80%.


at the same time, the Encyclopedia of knowledge map Sogou, "friend" can be accumulated through processing, engine technical resources, a better understanding of the user in the natural expression of communication. For example, the user said Tsinghua Qiu Yong, "friend" can use the knowledge map to know what he said is what Qiu Yong.


not only that, "bosom friend" and also users conducted several rounds of dialogue, user interaction logic processing more complex, interacting with a more natural and easier for users to accept and understand the way, step by step positioning the end user demand point, so as to provide more humane and convenient service. For example, when a user sends a request to the capital airport, "bosom friend" will further questions, you go to the capital airport T1 or T2, to determine when the user terminal, it will ask of or access. This is because Sogou can map the data structure, and through the surrounding landmark completely into a net, to such a depth data in speech, it can really achieve a more natural interaction with the user. />



, the industry believes that with the current speech technology on the market, Sogou "friend" is starting from the massive real user data, combined with the vertical scene to solve the user pain points, not only greatly improves the user's voice interactive experience, let the interaction mode become more natural. On the other hand, the "salon" engine will also promote the technology innovation of the iterative intelligent voice industry, as Sogou in natural interaction and knowledge computing progress, and industry partners will polish speech technology better and more perfect, I believe the future of speech interaction experience will enhance the quality and leap.


! Wang Guanxiong, the famous Chinese observers, ten from the media (see the authoritative list). Hosted and participated in the 4 IPO, the traditional enterprise transformation "Internet plus" coach. A depth of the daily article, published in WeChat, micro-blog, search engine, the major portals, technology blog and other nearly 30 mainstream platforms, covering 4 million of China's core business, technology groups. For the financial times, Forbes and other world-class media writers, the view is widely reproduced by the media quoted a great influence. For more details, Baidu, business consulting 13521140909, QQ1550505421.


RGB (0, 176, 80); font-family:; font-size: 15px; font-weight: bold; line-height: 24px; text-align: center; whit E-space: pre-wrap; background-color: RGB (255, 255, 255); "> Olympic marketing


" original "

; Wang Guanxiong (wang-guanxiong)

The lastest articles of wangguanxiong

The Spring Festival is not closing, consumers say is really good

Inventory of 2017 major companies annual meeting of the most in the three,...

No play devil trade war, this program is really the business guide

Consumer Internet dividend disappeared, creating a business model can open...

Multitray "double eleven": 24 hour combat notes

[my boy's school] with the power of faith

Women's volleyball team spirit, the spirit of the Internet and the 360 spirit