Potok or Molvi: which to pick for Russian speech

Potok transcribes Russian speech; this site has been translated into English for convenience.

Potok and Molvi are both voice input for Windows, macOS and Linux, but they differ on price and privacy. Molvi is free forever, with no account and no payment, and recognizes speech locally on the device: faster on an NVIDIA GPU, slower on a CPU. Potok (potok.now) is a cloud service with a model built for Russian speech and a free limit of 2500 words per week, then a paid subscription.

Molvi and Potok solve the same task, voice text input, but people choose them for very different reasons. Molvi appeals because it is free with no conditions at all: no account, no card, the program just works. Potok is paid beyond its free limit, but it has a model trained specifically for Russian speech and professional vocabulary. Which one wins depends on what matters more to you: price or specialization.

Where Molvi is strong

Molvi's main advantage is price, or rather the absence of one. The program is free forever, with no account and no time limit on use. The developer accepts only voluntary donations through Telegram Tribute, and they do not affect which features are available. At most competitors the free plan is capped by minutes or by words per week. Molvi has no paid plans at all. This is not "free with caveats"; there is simply no paid line.

Second, offline recognition that adapts to the hardware. On a computer with an NVIDIA graphics card, Molvi uses a higher-quality model, and on an ordinary CPU it switches to a fast, lighter version. The program adjusts to the capabilities of the specific device instead of demanding minimum system requirements up front.

Third, language. Molvi supports Russian and English with automatic language detection: you do not have to switch the recognition language by hand before each dictation if your speech is mixed. The site does not mention a separate personal dictionary for terms and names, so the model recognizes specific vocabulary as it saw it during training, without targeted tuning.

Payment and where data is stored

The comparison is simple here: Molvi has no payment at all. The program is free forever, and you can support the developer with a voluntary donation through Telegram Tribute, if you want to and not because you need to. Potok's free limit is 2500 words per week, and beyond that you need a paid subscription, paid in rubles with a card from a Russian bank. The legal entity behind Molvi is not named in open sources. For a free program with no account this is not unusual, but keep it in mind before trusting the program with work texts.

Data is as simple as it gets with Molvi: recognition runs entirely on the device, and audio is not sent anywhere, neither to the developer's server nor elsewhere. For people who insist that their voice must physically never leave the computer, this is the most direct answer among the services compared here. A similar approach is covered in the comparison with GigaType.

With Potok, data always goes to the cloud, although to servers in Russia. Recognition there does not depend on the specs of the user's computer, but it needs an internet connection and, beyond the free limit, a subscription.

How each handles Russian

Molvi supports Russian alongside English with automatic detection: the program works out which of the two languages you are speaking and does not make you switch modes by hand. The developer does not publish data on the stated accuracy for Russian speech specifically. The service is young, and there are few detailed technical specifications of the model in the open.

Potok runs on its own model trained specifically for Russian speech: names, professional terms, mixed Russian and English text. In the internal GolosTest, Potok has a WER of 2.1%. This is our figure from our own test, and it cannot be compared directly with Molvi, because there is simply no comparable accuracy data for Molvi.

For conversational speech and simple messages the difference is probably barely noticeable. For professional terminology, the model's specialization usually starts to matter, but you can only check that on your own text.

Sources: molvi.tech · vc.ru

When it is better to stay with Molvi

It makes sense to stay with Molvi if price is the deciding factor and you are fine with offline recognition without guaranteed high accuracy on complex terminology: the program is free, works without the internet and asks for no commitments.

Keep Potok's architecture in mind too: it is a cloud service, recognition runs on servers in Russia, and it does not work without the internet, unlike Molvi, which is built for offline mode by design. If audio must not leave the computer at all, this difference settles the choice against Potok.

Those who are already used to Molvi and its free model also have no reason to rush to a paid service for features they do not use every day.

Frequently asked questions

Primary sources: molvi.tech · vc.ru