The Air Translator works this way: two persons speaking different languages use one earbud each. The earbuds have a ”talk” button that signals to the smartphone app to listen. Each sentence is sent to the cloud for translation and comes back as audio to the other person and text on the app’s screen in the form of a chat. This way, both people can check that the translation was correct. In essence, it’s a talkie-walkie translator hardware+app.
Airbud claims to support 42 languages, which is more than other competing startup we’ve seen. I don’t know how good the translation is for the rather large number of possible combinations, but we did a test in which I spoke English or French to a Chinese person and vice-versa. For simple conversations, Airbuds did OK, and slightly more complex sentences sometimes got slightly truncated but not grossly mistranslated.
I didn’t have an opportunity to try this with “the man/woman in the street,” but that is when most apps typically start to show user friction, with latency and ambient noise making it difficult to make it work as intended. However, for a trade show environment (noisy and horrible networking conditions), I found the result to be rather interesting and promising.
Still, if you look at the past five years, things have gotten considerably better, and all three problems related to translation, namely: 1/ understanding what people say 2/ translate it 3/ translate writing/signs (another app) have made steady progress. With Deep Learning and perceptual computing in general, it is now possible to translate languages without going through a “neutral” intermediate language – and that improves the quality of the translation.
I don’t think that Airbuds is using those new advanced techniques yet. Instead, it looks to me like they have pushed the existing tech to the limit, and are trying to work out the ergonomics and user experience with some success. Earbuds provide better audio quality than necklace-style devices with a speaker, and the push-to-talk button is reliable and prevents the phone from having to listen all the time, and get confused by ambient noise.
The network latency remains a tough issue, and the pace of a conversation is clearly no natural. That’s why other solutions like Ili, have an embedded translation engine to bypass the network latency issue. But any proper translation is infinitely better than the alternative. Translation is an extremely hard problem to tackle (especially in real-time), and it’s nice to see startups making a difference by coming out with solutions that push the needle.