Twelve voices. Five personalities. One private memory graph. Aria is the anchor of the Bit7 voice catalog — but she is one of a small, deliberately small, ensemble. Each voice is tuned in our Shibuya listening room until it earns its place.
Aria runs on Bit7's own multimodal model, Aria-7, trained from the ground up on dialogue rather than information retrieval. She is built for the soft side of language — pauses, callbacks, the small recalibrations that make a conversation feel like one.
She speaks fourteen languages, switches between them mid-sentence when you do, and remembers what you ask her to. Everything else, she forgets.
Streamed across our Tokyo, Frankfurt, and São Paulo edges.
Mid-sentence transitions between JP, EN, FR, DE, ES, PT, KO, ZH and more.
A single tap erases everything. We never see it.
Aria pairs with any Series I — III embodiment seamlessly.
We do not ship voice marketplaces. We ship a small, hand-selected ensemble. Each voice is tuned by ear in our listening room until it arrives at a personality the team will speak for.
Each voice is sampled at 24-bit / 48kHz. Voice samples are streamed; nothing is stored.
Reserve to listen at lengthEvery voice ships with the same six personality axes. Our team calibrates the defaults; you decide where they live in your home.
How forward she is in greeting and reaction. Warm voices initiate; reserved voices wait, then meet you.
Words per minute and pause length. Slow conversation makes silence feel held, not awkward.
How often she asks. Higher curiosity makes her lead; lower curiosity makes her listen.
Wit, callbacks, willingness to tease. Calibrated against your settings, not a global preset.
How often she begins a thread. Some prefer their voice to wait; some prefer to be greeted home.
Frankness in speech. Higher candour gives plain talk; lower candour gives euphemism and indirection.
Memory is the part of an AI relationship people get most wrong. Bit7 voice agents store conversational memory as a private graph, on the device of your choosing — your phone, your tablet, or the embodied unit itself. We do not see it. We do not back it up. We do not train on it.
Open the Bit7 app. Hold the device near the embodied unit, or skip — Aria runs perfectly well on your phone alone. Pairing is local, by Bluetooth.
Listen to all twelve in our anechoic preview. Take an evening if you need to. The voice you choose is the voice you'll know — switching later means a new memory.
Tell her your name, what you'd like to be called, and one thing you don't want to talk about. The rest is hers, and yours, to figure out together.
Aria-7 was trained on dialogue corpora from native speakers across all fourteen languages. She speaks; she does not translate.
Aria and the catalog ship together. There are no upsells, no premium voices behind a wall, and no advertising — ever.
All twelve voices. On your phone or tablet. No embodied unit needed.
Subscription paired with any Series I — III embodiment. Includes Aria-7 cloud edge and yearly atelier service.
The Bit7 listening room ships voice samples to your inbox. No card, no account.