GigaType: what it is and how it works

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

GigaType is a voice input program for Windows and macOS that recognizes speech offline, right on the device, using Sber's GigaAM neural network trained specifically for Russian. The site has no public data on the price, the free plan or the developer's legal entity. Potok (potok.now) is a cloud alternative with open plans: 2500 words per week for free.

GigaType is a desktop voice input app that is built not on its own from-scratch development but on a ready speech recognition model, GigaAM, a neural network from Sber trained specifically for Russian. It works offline: the sound does not leave the computer. At the same time the developer's site has almost no open information about the price, the free plan or the legal entity, only the fact that the program exists and its technical basis.

How GigaType works

The program is installed on the computer and recognizes speech locally, without sending audio to a server: all processing runs on the GigaAM neural network developed by Sber and trained specifically for Russian, including its characteristic sounds, stresses and turns of speech.

GigaAM was not created for GigaType specifically, but as a general-purpose speech recognition model that developers of third-party programs can build into their apps. For the end user this means that recognition quality rests on the work of a large company, and not only on the efforts of the small team that makes the GigaType program itself.

Offline mode here is not an extra option but the only stated way of working: unlike programs with a hybrid approach, where you can choose between a local and a cloud model, GigaType, judging by open data, always works on the device. This also determines the computer requirements: a local neural network computes on the user's processor or graphics card, not on the developer's servers.

The developer's site gives no details about a personal dictionary, support for languages other than Russian, hotkeys and text insertion into specific apps: only general information about the technical basis of the program is available. Those who need such details before installing will have to either try the program directly or write to support with a question.

Sources: gigatype.app

Who GigaType suits

• Those for whom it is essential that the sound does not leave the computer at all: offline mode by default solves exactly this task, with no exceptions and no switches.
• Those who want a model trained specifically for Russian, and not a general-purpose multilingual neural network that splits its attention between dozens of languages at once.
• Those with a modern, powerful computer: local recognition loads the device more than cloud recognition, and on a weak machine it can run noticeably slower or use more battery on a laptop.
• Those who are ready to use the program without clear information in advance about the price and terms: you will have to either try it for free, if that is possible, or ask the developer directly before counting on anything.
• Those who already use other products built on the GigaAM model or simply trust Sber's development in speech recognition more than models with no publicly named origin.

Limitations of GigaType

The main limitation is transparency. There is no open information on the developer's site about the price, the free plan or the payment method: no price list and no description of a free version, if one exists at all.

The legal entity is also not named explicitly on the site: who exactly sells and supports the program could not be established from open data. For personal use this is not always critical, but for a business that needs closing documents it is a reason to clarify everything directly before paying, if a paid version is provided at all.

The only source of information about the program is its own site: in the course of checking, no independent reviews, comparisons or user feedback that could be cited were found. This does not mean the program does not work as stated, but it is harder for an outside observer to verify the promises than with better-known services. Other voice input options are collected in the section voice input programs.

When Potok is not a replacement for GigaType

Potok will not replace GigaType where offline mode is essential: GigaType, apparently, always works locally, and Potok never does. Potok is a cloud service, recognition runs on servers in Russia, and it does not work without the internet.

For those who care most about maximum transparency of terms before paying, this is easier with Potok: plans, limits and the payment method are open on the site, unlike GigaType, where part of this information could not be found. A detailed comparison of the two programs is in the article GigaType or Potok.

Frequently asked questions

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