Potok or GigaType: which to choose for Russian speech

Potok turns speech into text in 33 languages, including English and Russian.

GigaType is a dictation program for Windows and macOS with fully local recognition: the sound is processed on the device by Sberbank's GigaAM neural network, without an internet connection. Prices, the free plan and the payment method are not published on the developer's website. Potok (potok.now), on the contrary, works through the cloud: recognition runs on servers in Russia, and the free limit is known exactly, 2500 words per week.

GigaType and Potok bet differently on what matters most: where the voice is physically processed. GigaType keeps everything on the user's computer and at the same time publishes neither the price nor the payment terms on its website. Potok is the other way round: cloud-based, but with fully open terms from day one: plans, limits and the payment method are visible right away. The choice here is not only technical but also about how much information the service is willing to show in advance.

Where GigaType is strong

GigaType's main advantage is recognition entirely on the device. The GigaAM neural network, developed by Sberbank specifically for Russian, runs right on the user's computer: the sound is not sent anywhere, and without an internet connection the program does the job just as well as with one. For tasks where sound leaking outside the device is unacceptable in principle, such as confidential negotiations or work with sensitive documents, this is a structural advantage over cloud services, not a matter of privacy settings.

The second is the model itself. GigaAM was created specifically for Russian speech rather than adapted from an English or multilingual base: for terms, names and colloquial expressions, a model whose native language is Russian usually gives more predictable results than a model originally trained on another language and fine-tuned for Russian later.

The third is platforms. GigaType works on both Windows and macOS: those who switch between systems do not have to look for a separate solution for each.

Payment and where data is stored

Here the programs have opposite transparency models. Potok publishes its plans, limits and payment method openly: 2500 words per week for free, 14 days of full access without a card, then a subscription in rubles with clear terms.

With GigaType it is the other way round: neither the price, nor the free plan, nor the payment method from Russia is published on the developer's website. The legal entity is not stated in open sources either. For those who choose a program on the principle of “first I understand the terms, then I install”, this is a noticeable inconvenience: the real cost can apparently only be found out after installing or contacting the developer directly.

With data the situation is reversed: GigaType does not send sound to a server at all because it recognizes it locally, while Potok always sends data to the cloud, albeit to servers in Russia rather than abroad. For some, local sound processing is the main argument for GigaType; for others, a predictable price matters more than the route of the sound.

How each copes with Russian

For GigaType this is a fundamental architectural choice: the program is built on GigaAM, Sberbank's neural network trained specifically for Russian speech rather than adapted from a general model. The very name and origin of the model are already a sign of specialization.

Potok is trained on its own model, separate from GigaAM and any third-party developments, also specifically for Russian speech, including names, terms and mixed Russian and English phrases. On the internal GolosTest test Potok has a WER of 2.1%; this is our number from our test, no independent comparisons of recognition accuracy between GigaType and Potok exist, so it is impossible to say directly whose model is more accurate.

The difference lies more in the origin of the model than in measurable quality: GigaType has an open development by Sberbank, Potok has its own closed model.

Sources: gigatype.app

When it is better to stay with GigaType

GigaType remains the logical choice where offline operation is not a wish but a hard requirement: the sound must stay on the device under any circumstances. Potok does not meet this requirement in principle: it is a cloud service, recognition runs on servers in Russia, and it does not work at all without the internet.

It is also worth staying with GigaType if you are fine working without clear information about the price, for example if the program is already installed and working and the question of cost is not essential yet. As soon as it becomes important, the difference from Potok's open plans will be visible right away: 2500 words per week for free and a clear subscription after that. Other comparisons of this kind are collected in the overview of voice input software.

Frequently asked questions

Primary sources: gigatype.app · support.microsoft.com