iFlytek XiaoYi / iFlytek Translator

What it is

iFlytek Translator (讯飞翻译机), first sold as Xiaoyi Translator (晓译), is a handheld machine translation product from the Chinese company iFlytek. A user speaks into the device and it returns a translation in another language, so that two people who do not share a language can hold a conversation. It is a consumer hardware product built on the company's machine translation research.

Where it came from

iFlytek's own brand history traces the product line back to February 2012, when the company set up a machine translation research effort ^[1]. The first product, Xiaoyi Translator 1.0, was formally released in July 2016, and the brand history describes that release as creating the AI translator product category ^[1].

The company's accounts do not agree on every date. A separate iFlytek news page places the hardware translator in November 2016, a few months after the July date given in the brand history ^[2]. Both pages are published by iFlytek itself, and readers comparing them will find the two dates side by side.

Adding offline translation

The next step was to make the device work without a network connection. The brand history says a second-generation Xiaoyi with offline translation was first released in November 2017 ^[1], while the news page says the company launched an offline neural machine translation engine and introduced the first offline translation feature in September 2017 ^[2].

The term matters. Neural machine translation means the system uses a neural network to produce a translation for a whole sentence at once, rather than cutting the sentence into phrases and swapping them one by one. Running such a model offline means the computation happens on the device itself, with no connection to a server, so translation remains available in places where there is no usable network.

Translator 2.0 and the four standards

iFlytek Translator 2.0 was released in April 2018. With it the company proposed four industry standards for this kind of product: hearing clearly, understanding, translating accurately and pronouncing beautifully ^[1]. The four phrases describe a chain — clear audio capture, correct recognition of what was said, an accurate translation, and natural-sounding speech output.

The news page gives the capability in numbers: Translator 2.0 supported instant translation between Chinese and 33 languages ^[2]. A December 2018 update raised the number of languages to 51 and added industry-specific AI translation ^[2] — that is, translation tuned for the vocabulary and set phrases of particular industries, where ordinary everyday language models tend to be weakest.

Translator 3.0

Translator 3.0 followed in May 2019, with a new translation engine that the brand history says achieved 0.5-second translation ^[1]. The figure refers to the gap between finishing a sentence and getting the result, which is what decides whether a conversation feels like a conversation or like waiting for a machine.

A public role

iFlytek says the product was the official designated translator of the Boao Forum for Asia in both 2018 and 2019 ^[1]. An international forum of that kind is an unusually demanding setting: many languages in the room, background noise, and no tolerance for a slow answer, which makes it a visible test of the accuracy and speed the company had been claiming.

Where it sits in the line

Read together, the releases show a product moving through several stages: research begun in 2012, a first consumer device in 2016 that the company presents as the start of a category, offline translation in 2017, a 33-language 2.0 in April 2018, 51 languages plus industry-specific translation later that year, and a 0.5-second engine in 2019. Two themes run through it — more languages, and less dependence on a network — alongside the effort to state what a good translator device should do at all, which is what the four standards were for.