Voice OS

An operating system is the base software of a device. It launches programs, manages files, and accepts the user’s input. On a phone, this has so far happened mainly via the screen: tapping, swiping, typing. A Voice OS shifts this access point to the voice. The user says a sentence like “Send Mia the photos from last night,” and the system handles the individual steps itself. The screen doesn’t disappear, but it becomes secondary.

Why voice suddenly works as a control method

Voice control has existed for over ten years. Siri, Alexa, and Google Assistant, however, could long only handle fixed commands. Anyone who phrased a sentence slightly differently got an error message. These systems worked like a vending machine with fixed buttons: setting an alarm, yes; anything unusual, no. That’s why voice remained a niche function for the kitchen and the car.

With modern language models, this has changed. These are programs that have learned from vast amounts of text how language is structured. They also understand awkwardly phrased sentences and can sort out multiple instructions within a single sentence. This makes voice, for the first time, robust enough to control an entire device rather than just a single function.

For tech companies, a lot of money is at stake here. Whoever controls the operating system controls access to all apps. So far, Apple and Google share this market for phones almost entirely between them. A Voice OS is an attempt to attack this position, because with voice control, the app screen falls away as an organizing principle.

From spoken sentence to executed action

The process consists of several stages. First, speech recognition converts the audio recording into text. Then a language model checks what the user actually wants. Next, the system selects the appropriate function, such as the calendar module or the camera. Finally, a response is generated and read aloud.

The crucial part is the third stage. The model must not only talk, it must be allowed to act. For this, it is given access to so-called interfaces, i.e. defined connections through which programs give each other commands. Through such an interface, the system can send a message or create an appointment. You can think of it like an assistant who not only listens but also holds the keys to all the cabinets.

This is exactly where the biggest problem arises. Language models occasionally invent things that aren’t true. In a chat, that’s annoying; in a bank transfer, it would be costly. That’s why developers build in follow-up questions before sensitive actions are carried out. A second issue is latency: if there’s more than about a second between question and answer, the conversation feels unnatural. Many systems therefore process simple requests directly on the device and only send complicated ones to a data center.

Where Voice OS already exists today

The most visible examples are new devices without a classic screen. The Rabbit R1 and the Humane AI Pin were the best-known attempts, but both failed with testers. They were too slow and could do less than a normal phone. OpenAI is working with former Apple designer Jony Ive on a device of this kind. Such reports regularly appear in the business section, because they involve billions of dollars.

More often, you encounter the technology unobtrusively within existing products. Apple and Google have upgraded their assistants with language models so that they can work across apps. In cars too, voice control is increasingly replacing nested touch menus. The term Voice OS describes more of a goal than a finished product: a device that can be operated entirely through voice alone. How far this goal will carry is uncertain — on the bus or in the classroom, typing simply remains more practical.