Timekettle WT2 Plus: Fascinating translation headphones, but dangerous?
19.6.2019
Translation: machine translated
Video: Stephanie Tresch
The Timekettle WT2 Plus translates from 20 languages into any other language. Live. They are a really exciting piece of technology, but they raise a lot of questions.
Headphones that translate any language into any language. This is what the Chinese manufacturer Timekettle Technologies promises with the WT2 Plus. The in-ear headphones are somewhat reminiscent of Apple's AirPods. The case is a kind of shell that you crack open in the centre like an egg.
The design from the studio Innozen Design] has been honoured with the IF Design Award, according to the Timekettles website. The design jury praised the simplicity of the device. Well, if you're inspired by the AirPods, then "simple" comes automatically. However, you will still find some Chinese pomp, i.e. the kitschy, eye-catching overemphasis of individual elements: when you charge the WT2 Plus, the letters W and T embedded in the case flash like crazy. Less would have been more here.
But this is really only relevant until the battery is charged and the translation can start. And then, of course, the experiment to see how the device can be brought to its knees. So let's find people who speak a foreign language that the WT2 can't cope with
Privacy risk: Your words in the cloud
The WT2 Plus actually works quite simply from the user's point of view:
- The microphone in the earbuds records the speaker's voice
- The recording is sent to the cloud.
- The recording is translated. The WT2 Plus can translate over 20 languages and several dialects
- The headphones or speakers output the translation
This is just as simple but tricky when it comes to the safety side of things. Neither the package leaflet nor the Timekettles website says which jurisdiction your data is transferred to and who has access to it and under what circumstances.
The reason I'm mentioning this separately here is that the WT2 Plus lends itself to conversations in a business context. Conversations that don't necessarily have to go public. Theoretically at least, it is possible for a competitor to connect with Timekettle as an "agent" or "contractor" and then gain access to your data. If this is not the case, Timekettle is not doing itself any favours by not explicitly stating that it will never sell specific information from your conversations to third parties.
Furthermore, it is not entirely clear which data is stored where and in what form and which rights to the data flow where. Here too: Timekettle is not doing itself any favours by not explicitly defining this somewhere.
A small disclaimer here: I'm not saying that Timekettle sells the data to everyone or does shenanigans with it. I'm just saying that there is a theoretical possibility of fraud and a lack of public information.
Because not every conversation is a friendly chat in the street or an order in a restaurant.
The app in detail
I suspect that the hardware is not really necessary. The headphone hardware is possibly only there to standardise the sound quality of the microphone. If Timekettle were to rely on the on-board microphones of a smartphone, there would be too many variables that could influence the sound quality. Nevertheless, the main functionality of the WT2-Plus system is in an app (Android and Apple iOS) or the cloud.
The app has three modes:
- Auto Mode: The headphones are always listening and translate live
automatically2. Touch Mode: The headphones only listen when you press a contact surface on the side. The translation starts as soon as you let go
- Speak Mode: The person being translated wears one or both in-ears, presses a button on the app and speaks. When they release the button, the translation starts and is output via the smartphone speaker.
In Speak Mode, it is not absolutely necessary to put the headphones in your ears. In a small, quiet room, it is possible to place the WT2 Plus on the table and speak. The Auto Mode is quite sharp and sometimes translates sentences spoken in the other language. The results are highly amusing. Touch Mode is a healthy mix of both, but has delays in conversation.
The headphone app combination listens, sends the data somewhere - the translation engine used by the WT2 Plus is also not specified anywhere - and returns with the translated text within an impressively short time.
The song of a language
The quality of the translation is the same everywhere, but it depends on the language. In the vast majority of cases, the translation is somewhere between above average and certainly functional to almost perfect. The WT2 Plus has at least English, Spanish, French, Croatian, Russian and Turkish very well under control. On the other hand: The WT2 Plus falls down when it comes to the Tamil dialect spoken in India. The translation spoken by the robot voice has absolutely nothing to do with what is supposed to be communicated.
A guess: The ability to understand and translate you depends, among other things, on the melody of the language to be translated. The more defined the singsong of the language, the easier the translation.
A comparison. The Russian language sounds like this:
Youtuber Evelina Barry talks about herself, her culture and her life. There are sibilants, defined A and E sounds, hard Ts and rolled Rs. It's easy for software, because the sound she records leaves few gaps.
In French, sounds are often swallowed, for example the E at the end, but the word itself remains the same. From a purely acoustic point of view, it doesn't matter whether you say "Ami" to your boyfriend or "Amie" to your girlfriend. At the very most, the app could then - if you don't include a gender-specific descriptor in the sentence - eject "My friend Anne-Sophie". But if instead of "Mon amie Anne-Sophie fait des choses" you say "Mon amie Anne-Sophie, elle fait des choses", then the app is able to take the gender-specific descriptor "elle" as the gender of the entire sentence construct.
Tamil then, now without the Hindi inflection:
The fricatives in particular - in simple terms, F sounds - are sometimes swallowed, sometimes not. Although this creates its own melody, it means that the software has to take a further step towards abstraction. Because if a sound is not pronounced, it could be there, but it doesn't have to be. So the software has to think about this and be able to intercept it. However, since it doesn't know where the swallowed fricative is, it could theoretically be anywhere. Therefore, the app probably looks for the context and then tries to plug the holes in its understanding. The potential for error with this approach is of course high.
A conversation with the world, and everyone who buys your data
All in all, however, the WT2 Plus is a fascinating technology. Even if you're only speaking your secrets to the cloud and to "agents" and "contractors", the headphones work amazingly well. The world feels a little smaller, because during the test I spoke to a Vietnamese woman, a Russian woman and a Tamil man from India. It's fine, except that I can't be convinced of the security of your privacy.
So, done. By the way, Hollywood actress Jodie Foster speaks absolutely excellent French. I noticed this during my research.
Journalist. Author. Hacker. A storyteller searching for boundaries, secrets and taboos – putting the world to paper. Not because I can but because I can’t not.
Smartphone
Follow topics and stay updated on your areas of interest
Smartphone
Follow topics and stay updated on your areas of interest
11 comments
later