In the old story, humanity spoke a single language and started building a tower tall enough to reach the sky. The work was going too well, so the languages were scattered: suddenly nobody on the scaffolding could understand anyone else, and the tower was abandoned. The barrier has stood ever since.
Babele is named after that tower, because for the first time the scattering is reversible in software. Not "look up a word" reversible. Reversible as in: two people who share no common language can hold a real conversation, out loud, right now.
Generative AI quietly turned "I do not speak your language" from a permanent wall into a solvable problem. Babele is my attempt to make that concrete and put it in a pocket.
The promise, made concrete
Picture the actual situation. You land in a country whose language you do not speak. You need directions, or you want to ask the woman at the market what is actually in the dish, or you just want to talk to the person next to you on the train.
Today the options are grim: point at a phrasebook, or type into a translation box and rotate your screen toward a stranger like you are serving them a document. The conversation dies before it starts.
Babele is the other version. You put a Bluetooth earpiece in your ear, your phone stays in your pocket, and you just talk. The shopkeeper talks back. You hear them in your language, they hear you in theirs. It is a conversation, not a transaction with a machine in the middle.
The hardware bar is deliberately low. You do not need special glasses or a dedicated gadget. A twenty euro Bluetooth earpiece is plenty, and you probably already own one.
What Babele is
Two people, two languages. Babele listens, translates, and speaks. That is the whole product.
You pick your language and the other person's language, then you start talking. There are two ways to use it:
- Phone only. The phone is the interpreter. Set it on the table between you and both of you talk to it. No accessories required.
- With an earpiece. Pair any Bluetooth headset and the replies meant for you arrive privately in your ear, while the other person hears their side out loud.
No accounts, no sign-up wall, no setup ceremony. Pick two languages and go.
The detail that makes it feel natural
The thing that turns this from a gimmick into something you would actually use is who hears what.
When you speak, Babele plays the translation on the phone speaker, so the other person hears it. When the other person speaks, the translation comes back privately in your earpiece. Nobody has to share an earbud, and nobody has to listen to both halves of the conversation twice.
For that to work, Babele has to know who is talking at any moment. It does not ask you to tap a button or hold a key. It detects the spoken language on the device itself, turn by turn, and routes the audio to the right place automatically. You speak your language, it goes out to them. They speak theirs, it comes back to you. You never think about it, which is the point.
Phone mode mid-conversation. Your turns on the left, the other person's on the right, each bubble showing what was said and the translation underneath.
How it works, briefly
Under the hood it is simpler than it sounds. Audio streams to Google's Gemini Live API, which does speech to speech in real time: it hears one language and speaks the other, with a natural voice, while the conversation is still happening. The "who is speaking" decision uses on-device language identification, so the routing choice never leaves the phone.
There is no Babele server in the middle. The app talks directly to Google's API and to your earpiece, and that is the entire backend. The longer technical write-up, including the audio routing and the latency work, lives in the project's docs.
Bring your own key
Babele does not ship with an API key, and that is on purpose.
The first time you open it, it asks you to paste your own Gemini API key. Getting one is free and takes about a minute, and the app walks you through it. From then on the key lives only on your device, it is sent only to Google's API, and usage is billed to your own account.
This is the honest version of a personal AI tool. There is no shared key to run dry, no quota I have to police, no incentive for me to collect your conversations to subsidize a free tier. You bring the key, you stay in control, and I never sit in the middle of what you say. Babele is open source under the MIT license, so you can read every line that handles your audio and your key.
And yes, it works great with smart glasses
If you do happen to own a pair of Ray-Ban Meta, they are the nicest version of this. The microphone and the open-ear speakers are already on your face, so the whole thing becomes truly hands-free: you look at the person, not at a phone, and the translation just happens.
But that is a luxury, not a requirement. The glasses are simply a very comfortable Bluetooth earpiece. Any earpiece does the same job, which is the part I want to be loud about: this is not a four-hundred-euro accessory feature, it is a phone app that happens to be lovely with glasses.
Honest limitations
It is early and experimental, and it behaves like it. The earpiece has to be paired as a normal Bluetooth headset first. Audio over that Bluetooth voice channel is narrowband, so the voice in your ear sounds a little telephone-ish. Very short utterances, a single "yes" or "okay", can be ambiguous for the language detector, and the routing occasionally guesses the previous direction. None of this breaks the experience, but you will notice it.
Try it
Babele is free and open source. If you travel, or just want to talk to someone you currently cannot, point it at your next conversation.
Feedback, especially from real conversations in the wild, is very welcome.