The company
About VoiceFrom.
In short
VoiceFrom makes computer programs for speaking and listening. They listen when a person speaks. Then they say the same thing again, in your language or with easy words. So you can understand people in class, at work, or on a podcast. VoiceFrom is a speech AI company. Our technology helps every listener understand what people say. It translates and simplifies speech, live, to match the listener's language, how well they speak it, and how much they already know. It works in conversations, lessons, meetings, and podcasts. VoiceFrom is a speech AI company building a world where no listener is left behind. It develops real-time technology that adapts speech to the person listening: to their language, how well they speak it, and how much they know about the subject. It works in conversations, lectures, meetings, and podcasts. VoiceFrom is a speech AI company building a world where no listener is left behind. It develops real-time speech intelligence that adapts whatever is being said to whoever is listening: to their language, their proficiency in it, and their expertise in the subject, in a conversation, a lecture, a meeting, or a podcast.
Same meaning, simpler words. Pick a level.
The problem
When we speak, the same message reaches very different listeners. Think of a language learner trying to grasp a popular podcast, a patient struggling to comprehend their doctor, a diverse audience speaking a dozen different languages, or a highly confidential meeting that requires secure, local processing. Today, the burden of comprehension falls entirely on the listener. VoiceFrom flips this dynamic. We build real-time speech-to-speech AI based on one core idea: the speech itself should adapt to whoever is listening, live, while still sounding exactly like the original speaker.
The products
The answer.
Each product adapts speech to one kind of listener, live, and keeps the speaker's own voice. Accuracy and latency are scored in the open: the harness and the method are on the blog. Interpret Cloud and Restate can be heard on this site, on a live broadcast as it airs.
For eventsWe run it
Interpret Cloud
Interpret Cloud runs a multilingual session live: one source feed from the stage or the stream in, the audience's languages out, each speaker still in their own voice, straight to every listener's browser. Nothing to install, no headsets to hand out. When the microphone changes hands, the interpretation changes voice with it. For in-person, online, hybrid, and broadcast sessions.
It is tuned for meaning rather than word-for-word output. In VoiceFrom's head-to-head benchmark of April 2026 (five systems, the same eight sessions in English, Spanish, French, German, and Portuguese, scored the same way), VoiceFrom Interpret was among the most accurate of the five.
For closed networksYou run it
Interpret Air-Gap
Interpret Air-Gap delivers the exact same models and AI quality as Interpret Cloud, but runs entirely on hardware the customer controls. It brings cloud-tier performance directly to your premises, functioning entirely without an internet connection or cloud fallback. Nothing is uploaded, zero data leaves the room, and the entire system is fully auditable on-site. It keeps right on interpreting, even if you sever the network cable.
It is built for governments, parliaments, courts, regulated industries, and field operations handling highly sensitive environments: closed hearings, classified briefings, board meetings, and secure negotiations. Deployments are planned directly with the customer's security team, rigorously tested on their domain content, and signed off in a fully disconnected state.
Packets to the public internet · 0
For learnersYou set the level
Restate
Restate listens along and repeats what is being said at the listener's level, in the speaker's own voice, a few seconds behind. Same language in and out, simpler words, same meaning. Listeners can choose their preferred comprehension level (CEFR A2, B1, or B2), adjust it mid-stream, and follow along with live captions.
It works with live conversations or alongside any other app playing audio, whether it's a podcast, a video, or a lecture. It simplifies the language, but always keeps the tone and the voice of the original speaker intact.
Restate is built for learners, newcomers, and anyone navigating a second language, as well as the teachers, creators, and schools who support them. Launching first in English for the browser, with more languages and a mobile app on the way. We are currently opening Restate to early-adopter teachers, creators, and schools as founding partners to help shape its future.
Patent pending · PCT/IB2026/057029
People
The founders.
CEOFounder
Dominik Roblek
Seventeen years at Google, most recently leading the Sense Audio team in Google Research, which shipped Now Playing on Pixel and noise cancellation in Google Meet. Co-author of Now Playing, Fréchet Audio Distance, SEANet, SPICE, and AudioLM; more than 100 publications and patents in machine learning, including 78 granted US patents.
CTOFounder
Hassan Rom
Fifteen years of applied ML at Google: Search, Shopping, perception for Waymo and Image Search, and on-device audio understanding for Android, Pixel, and Nest. Co-creator of the Open Images dataset.
Brand
The name, the logotype, and the colors.
The mark, the wordmark, and both lockups: SVG for print and design tools, PNG at 2400 px wide for everything else. If you take one file, take the horizontal lockup in gradient and ink. Use the artwork as it is; please don't alter it.
The name
VoiceFrom, one word with a capital V and F. The company is VoiceFrom Inc. in legal copy and VoiceFrom everywhere else. The interpretation products are Interpret Cloud and Interpret Air-Gap; the benchmark posts call the interpreter behind both VoiceFrom Interpret. A product is VoiceFrom Restate on first mention, then Restate. The name reads as a phrase: the voice comes from whatever follows it, AI in the domain name, a product in VoiceFrom Restate.
Horizontal lockup
Stacked lockup
Mark
Colors
Coral · #E7553E
Ink · #1B1814
Off-white · #F9F7F5