With the development of economy and the improvement of living standards, it has become more and more common to talk about business and travel abroad, but after all, it is not domestic, and the language barrier has always been a problem for everyone. However, with the breakthrough of artificial intelligence technology, a new category - artificial intelligence translation machine has emerged this year. Xiaoyi, who was praised by the Prime Minister before, and Jingdong crowdfunding more than six times later. What's their actual performance? Is it satisfactory?

Appearance details

The appearance of iFLYTEK's Xiaoyi is quite different from that of the crossover tower. The Xiaoyi we just got is a black version, which is very sharp in contrast to the main color of the white version. It gives the impression that Xiaoyi tends to be bold and wild, while the Zhu'er is fashionable and compact. After a closer look, the smooth curve of Xiaoyi as a whole is more prominent, and Zhuan'er just makes a transition at the four corners.

The front design of Xiaoyi is obviously different from that of Xiaoyi. Xiaoyi adopts a symmetrical design, with control buttons arranged in the center, and two loudspeakers arranged up and down; The focus of Zhun'er is the LED touch screen, and the control buttons are arranged below. Maybe it's the difference in thinking, so Zhuan'er has more advantages in volume than Xiaoyi.

On the left side, Xiaoyi has set up a volume key, and Zhunyer is a very clean metal side.

Both on the right side are equipped with power keys, and there is a SIM card slot below Xiaoyi, which is an additional network connection mode than Zhunyer.

On the top of the machine, Xiaoyi is equipped with a microphone and a 3.5mm headphone port. Zhunle is only a microphone. What's surprising is that Zhunle has designed a unique lanyard hole.

The bottom is of course a familiar microUSB interface for charging. Because of the symmetrical design, Xiaoyi also has a microphone at the bottom, and Zhuoye places the speaker here.

Different people have different opinions on the appearance and style, and the better one is related to personal aesthetics. As for the workmanship, both of them are excellent, but the metal material used by Zhun'er in the back is more textured than the plastic body. At the same time, it is equipped with LED touch screen to make Zhun'er more comfortable in independent use. You can directly calibrate the content without turning out the phone; Xiaoyi's separate volume key and supporting SIM card also make the operation more convenient, but in terms of volume, it is really a failure.

Practical drill

For users, the core appeal of using AI translation machine is the accuracy of translation and the richness of languages. AI translation machine involves a number of cutting-edge AI technologies, mainly including speech recognition, natural language understanding, machine translation and speech synthesis. Among them, machine translation technology is the core, which is also the most difficult to overcome, and is known as the "pearl on the crown of AI".

Both iFLYTEK and Splitter Tower have made great efforts in product technology, and each has its own strengths. IFLYTEK is currently the best in Chinese speech recognition technology, while the splitter tower has more advantages in machine translation technology. What is its true performance? Can it meet our requirements?

Multilingual performance

First of all, in terms of language, Xiaoyi supports Chinese English translation. Zhun Er provides more language support than Xiaoyi. At present, Zhun Er also provides Japanese, Thai, Korean and Chinese translation in addition to Chinese English translation. In mid August, 18 more languages were launched through OTA update.

A comparison video is attached:

Translation accuracy performance

In order to avoid the bad influence of the author's accent on the results (In fact, it is because of poor foreign language skills) This time, we specially invited international friends from the United States, Britain, Australia and Japan with pure accents to help us with the test. We simulated ordinary daily conversations. The standard pronunciation of each language (for example, American, British and Australian were tested in English) was tested independently, and a total of more than 1200 sentences were tested. Let's see the actual results.

Sino British performance

The need to ask for directions and shop is inevitable when you go out. At this time, it is the turn of the interpreter. In our Chinese to English measurement, Xiaoyi and Zhun Er have basically perfect recognition of Chinese. Whether it is authentic Mandarin, Guangdong boiled winter melon, or even the unique accent of foreign friends, both can accurately recognize, and the measurement accuracy can reach more than 96%. In the whole recognition process, through the LED screen, we can see that there are occasionally a word or two errors in the Zhuier. Xiaoyi is unable to determine the specific recognition rate because the app does not support it.

The accuracy of actual translation, Xiaoyi slightly outperformed some, reaching 97%, and Zhun Er also reached 96%. There is no difference between the two. The final output of English results is generally one or two words different. There is no objection to the overall expression. Talking with foreign friends, they said that the results of translation can be understood. There is no major problem in communication through Xiaoyi and Zhun Er.

British performance against China

How did the UK perform against China? After all, when people are outside, there is nothing to rely on, leaving the translation function to build a bridge of communication. In the face of frank American accent or gentleman's English accent, Xiao Yi and Zhun Er can easily deal with it. As long as they are not particularly abnormal, they can both achieve more than 90% accurate recognition. However, Xiaoyi's accuracy rate in dealing with English accents has declined significantly, from 94% of American accents to 92%. Zhun Er's performance is relatively stable, reaching 96% and 94% respectively. The translation results are also satisfactory.

Even if it is quite wild, Zhun Er and Xiao Yi are still outstanding. What needs to be said is that Zhun Er seems to be better at English recognition. Although the gap is very small, it is just that Xiaoyi occasionally has wrong words, which makes people feel inferior. The specific accuracy rates are 95% and 94% respectively. There is no problem with the general idea of both when translating into Chinese. The details are more accurate and smoother. The accuracy rate of daily scene dialogue has reached 95%.

As for the problem of Chinese accent, after several poor attempts by the author, I found that Zhun Er and Xiao Yi had no pressure and showed no inferior performance. After all, they had conquered the pure foreign language accent, so it was natural for them to have only Chinese accent.

Performance of Sino Japanese Translation

In addition to the common Chinese English translation, Junger provides more language support than Xiaoyi. We mentioned earlier that Junger also provides Japanese, Thai, Korean and Chinese translation in addition to Chinese English translation. In mid August, 18 more languages were launched through OTA update. Here we also found a neighbor in the next 11 districts to take a Japanese test. Zhun Er's performance is still good, a little backward compared with the Chinese English translation. However, the translation accuracy of the Chinese Japanese translation and Japanese Chinese translation of overseas tourism scenes has reached 92% and 90%, which can meet the needs of daily communication.

tail

In this test, it can be said that the performance of the translator completely exceeded the author's expectations. The speech recognition introduced into the artificial neural network in 1960 has now been able to bite the hard bone of Chinese; With the explosion of big data, machine translation after the establishment of better learning models can make them no longer feel so rigid and cold. Generally speaking, this AI translator is competent as long as it does not involve communication in a very professional field. It is more than enough for daily conversation scenes.

Here is a simple science popularization. Generally speaking, speech recognition technology is that computers recognize human speech and convert it into characters; According to different languages, it can be divided into Chinese speech recognition, English speech recognition, Japanese, German, Russian and so on. And machine translation technology is that computers translate one language into another; Similarly, according to different languages, it can be divided into Chinese English translation, Chinese Japanese translation, Chinese German translation, Chinese Japanese translation, English Thai translation and so on. For AI translation machine, it is easy to understand, and machine translation technology is the key to it.

As a giant company of Chinese speech recognition, iFLYTEK is indeed in the forefront of the world in terms of Chinese recognition in speech recognition technology, which can be seen from Xiaoyi's Chinese to English results. But Chinese speech recognition ≠ machine translation, which is not the case for recognition of languages other than Chinese. Xiaoyi temporarily does not support translation between Chinese and English, which is also the case for translation introduced by iFLYTEK. It seems that English recognition is slightly inferior to Zhunge'er, which indicates that iFLYTEK has more room for progress. It seems that the voice splitter company also understands that Chinese speech recognition is not its own strong point, so it focuses on the core technology of the translation machine - machine translation technology. It relies on the deep neural network algorithm to carry out continuous optimization of oral translation, so that Zhunyer is slightly better than Xiaoyi in English to Chinese translation, and it is not difficult to understand that it supports more languages.

Returning to these two products, regardless of appearance, Xiaoyi and Zhun Er are on a par with each other in terms of the accuracy of important Chinese English translation. Both can freely talk with foreign friends, fully meeting the needs of people going abroad. In addition to Chinese and English, Zhunge'er also supports the translation between China and Japan, China and South Korea, and China and Thailand, which is more widely used. After this test, the author can't help sighing that the development of technology is changing with each passing day. With the development of artificial intelligence technology, it can be imagined that everyone can easily master dozens of languages through artificial intelligence translation machines in the future, which will have a profound impact on education, tourism, trade and other industries. Different countries and ethnic groups can better communicate and integrate. Will the dream of a global village be far away?

Follow our Weibo @ Love Computer

Follow our WeChat official account: playphone

Of course, we also pay attention to our Bilibili account: love computer