As a beginner in the new wearable translator area, we would like to share the knowledge that we know to you. If there are inappropriate expressions, welcome to any suggestions and advice.
We have communicated with lots of users via exhibitions, off-line meetings, on-line communities, etc. And we found three important points.
1. Wearable Translator Area is new, really new for us. Most people do not know much about wearable translator market, and their questions are very similar.
2.”Amazing!” “ Wow!” “Oh, really?” These are the first voice jumping from their mouth when they use our WT2 Real-time Translating Earphone. People like this translating technology, but has few chance to experience translating devices.
3.people expressed the desire on solving the language barriers when being stuck in a place without a means to communicate. It’s time to make machine translation be an accessible consumer electronics around you.
Here are 7 tips you should know.
1. Speech translation has 3 steps, ASR (Automatic Speech Recognition), MT (Machine Translation) and TTS (Text to speech). The whole process is Speech -> Text (One language)-> Text (Another language) -> Speech
2. ASR (Automatic Speech Recognition)
a. The fact you should know is that ASR is often treated as a very mature technology as lots of companies provide relative cloud service, and the accuracy of speech recognition is more than 95% in a quiet place. The key point of accuracy is up to the input sound quality.
b. Compared to noise, multiple mixing sound is one of the key difficulties for ASR. It’s difficult to recognize his/her voice when the device is far away.
c. In fact, the highest standard of ASR is about human cognitive abilities, but not the pursuit of 100% accuracy. As you know, it’s normal phenomenon that there is something wrong with the words or grammar even you both speak same language, because people communicate with each other by voice, body language, eye contact and their cognitive abilities.
d. ASR is a very complicated work which needs a vast quantity of computation. To ensure accurate results, it’s better to connect to the cloud, but not offline devices.
e. It’s easy to identify language from the text, but it’s difficult to identify language from a section of voice, especially different languages in the same language family. Due to the further development of globalization, languages are learning from each other, like 豆腐（doufu） and tofu，沙发（shafa）and Sofa.
f. It’s much more difficult to recognize the voice when speaking in multiple mixing languages.
g. Offline ASR (without internet) should be called the command word recognition, it means to match your words with the words in the offline database.
h. The more homophones have, the more difficult to recognize. For example, the Chinese homophones is much more than English.
3. Machine Translation Technology.
a. There are 1000+ kind of language in the world, that means there are 1000*1000 translating combinations. And there are lots of minor languages as well.
b. The way on how to translate some minor languages is to use English as a mediation. For example, the process of Korea to Chinese is Korea to English, then English to Chinese. The accuracy results are not very good generally.
c. It’s easier for the language translation in similar grammatical structures.
d. There is no absolute objective assessment on the accuracy of Machine Translation. The accuracy of machine translation from word to word doesn’t make sense in the accuracy of machine translation. However different people has different views on the standard of assessment.
e. To ensure the quality of accuracy, it’s better to connect the cloud for machine translation. And the quality of offline translation will sharply decline.
4. TTS (Text to speech)
a. TTS is the relatively easier technology among these three technologies. However, it is still difficult to reach the level of human.
b. TTS can be offline, but the quality is much worse than online version as well.
5. Few companies can do the work integrating all these three technologies, most companies just do one thing. And the algorithm has become more and more open, the core competence is the proprietary data, the more data in a scene segmentation, the stronger the training brain will be.
6. “Brain” likes the agricultural products, data likes fertilizer that you need to buy. In fact, most big companies would like to share free Open API to other companies, in order to get more “fertilizer” via different pipelines.
7. With the popularization and cost down of Machine Translation Technology, it will be the fundamental service like map technology, applied to various areas.
Finally, a brief introduction our product, WT2 Translator.
WT2, the real-time translating earphone for natural foreign communication, for true two-way conversations, the way people are used to connecting. Finally, foreign communication can be natural and fluent, preserving both eye and body language as speech is aided by the WT2 Translator. Imagine being able to travel the world without language barriers!
