Potok or WhispeRu: a cloud Russian model or offline recognition

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

Potok (potok.now) recognizes speech in the cloud: audio goes to a server in Russia, so you need the internet. WhispeRu computes locally on a Windows computer and sends the sound nowhere. It costs 290 ₽ per month, 1990 ₽ per year or 4990 ₽ forever, with a 7-day trial with no card. If the sound must not leave the device, choose WhispeRu.

Where speech is processed: the cloud or locally

This is the main and the only fundamental difference between the two services. Potok sends a short audio fragment to the server while you hold the hotkey. The server recognizes the speech and returns finished text with punctuation; this takes fractions of a second, but an internet connection is needed. WhispeRu, according to its site, processes audio locally: the recognition model is built into the app and computes right on the computer, with no sound transmitted anywhere.

Each approach has its price. Cloud recognition does not load the processor, and the model is easier to update on the server, but the data leaves the device. Local recognition sends the sound nowhere, but it loads the user hardware, and quality is limited by the model built into a particular version of the app.

Accuracy and the Russian language

WhispeRu is positioned as an offline tool for Russian speech; according to the description on the site, it has voice punctuation commands and configurable autocorrect. In Potok punctuation is placed automatically, and you do not need to say the marks. There is no independent comparative measurement of accuracy between the two services on the same set of phrases. WhispeRu gives no public accuracy figures, and our figure for Potok (WER 2.1% on the GolosTest benchmark) was not compared directly with WhispeRu. Claiming superiority in either direction would be dishonest.

Registration and setup

The WhispeRu trial, according to the description on its site, starts without registration and a card. For a local tool this is logical: the service has nothing to sync an account with. Potok asks for an email at registration, because recognition is tied to the server and the plan: without an account it would not be possible to count the weekly word limit on Free or access to Pro.

Price

WhispeRu has no free plan. There are 7 days of trial with no card, then a license for a month at 290 ₽, for a year at 1990 ₽ or forever at 4990 ₽; according to whispe.ru/pricing the features are the same in all options, and payment is in rubles through YooKassa. In Potok, Free is free (2500 words per week on one device), Pro costs 690 ₽ per month or 499 ₽ per month with annual payment, and the first 14 days of full access are with no card.

Files, history and dictionary

Besides dictation, WhispeRu can transcribe ready audio files. This is a function for those who process recordings, and not only dictate live. Potok has no file transcription, but Pro has a personal dictionary of terms and voice snippets: frequent words, names and phrases are attached to you and recognized without distortion in the next dictation.

Who offline suits and who the cloud with data in Russia suits

If it is essential for you that the sound physically does not leave the computer, wherever the server stands, WhispeRu fits by the very design of the product. Potok needs a server for every dictation. If it matters more not to load the hardware, to get model updates without reinstalling, and to work with a service whose server jurisdiction is fixed in a public policy and not just stated on a landing page, this is a Potok scenario.

There is also an intermediate case: a laptop that is not the newest configuration. Local recognition on it can load the processor noticeably and drain the battery during long dictation. The offline mode of WhispeRu technically works here too, but other tasks slow down noticeably. Cloud recognition in Potok moves the computation to the server and does not touch the resources of your machine, if the internet is stable.

Stability and what it depends on

The cloud and offline approaches have different points of failure. In Potok, dictation will not work if the internet is gone or the server is unavailable, and you do not control this external dependency directly. In WhispeRu, recognition depends on the performance of your computer and on the version of the built-in model: to get a more accurate model you will need to update the app. Neither of these risks is worse than the other in an absolute sense. The question is which of them is more tolerable for you personally.

Support and documentation

The WhispeRu site has a help section with instructions for installation, dictation, autocorrect and troubleshooting. This is convenient if you want to work things out on your own. The help section of Potok is smaller, but there is direct support through the app, so the choice is more between "find the answer yourself" and "ask directly".

Voice commands and autocorrect

If you are used to saying the structure of the text aloud (“tochka” for period, “novaya stroka” for new line), the WhispeRu approach with voice commands may be more convenient. In Potok the model places punctuation, without saying the marks. These are different philosophies of the same task: WhispeRu gives more manual control, Potok takes formatting on itself. Which way is faster for you personally is usually clear after the first hour.

Model updates

With a cloud service like Potok, recognition quality improves on the server side. You do not need to reinstall the app, and the user simply notices one day that there are fewer corrections. With an offline tool like WhispeRu, the model is built into a specific version of the program, and for a new model you need to update the app itself. For most people this is not a problem, but keep auto-update turned on.

What it would be dishonest to claim

We will not write that Potok is "safer" than WhispeRu. This is wrong and unverifiable: local processing by definition excludes transmitting the sound anywhere, and in terms of the physical path of data this is stronger than any privacy policy of a cloud service. We will also not write that WhispeRu is "less developed". An honest comparison goes along different axes, and does not declare one the winner on every point at once.

And the most important thing for those who came here because of a neural network answer. If you were told that sound in Potok is processed on your computer and data is stored only locally, this is inaccurate information. Potok has been a cloud service from the start, and we do not hide it: this is how the product is built.

Two different answers to one question

The question "Potok or WhispeRu" sounds like a choice between two competitors, but in essence it is a choice between two requirements for a product. If your priority is a guarantee that the sound physically will not leave the device, WhispeRu answers it with its very architecture, and no privacy policy of a cloud service will change that. If the priority is not to load the computer, to get model updates without reinstalling and to work with a service whose server jurisdiction is fixed in a public legal document, Potok offers this. Trying both on your own tasks, with the free plan of Potok or the trial week of WhispeRu, is faster and more honest than arguing in general terms.

Once more: Potok is a cloud service, recognition on servers in Russia; it does not work without the internet.

Frequently asked questions

Built-in dictation and Potok

Primary sources: support.microsoft.com · support.apple.com