Voice input for neural networks: how to use it

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

Voice input for neural networks is dictating prompts and questions instead of typing: a long request with the context of a task is quicker to say than to type by hand. Potok (potok.now) inserts the recognized text into any field, including the chat window of a neural network. By voice you get about 900 characters per minute against roughly 200 when typing.

You can dictate 900 characters per minute by voice against about 200 when typing; this is how the difference between dictation and a keyboard looks by Potok's figures. It is especially noticeable for long prompts to neural networks: a detailed request with the context of a task is easier to say in full than to type, stopping in the middle of a thought and losing the thread of the reasoning. This section collects materials on how to use voice for exactly this task: from single prompts to multi-step dialogues with clarifications and dictating code.

Voice input for neural networks: how people use it at work

The most common scenario is a long prompt with the context of a task: describe the situation, give the source data, formulate what is needed as output. Typing such a text is tiring and it is easy to lose your way in the middle; saying it aloud is usually quicker and more natural, especially when the thought has already formed in your head and you only need to pronounce it, without choosing words as you type.

The second scenario is a chain of clarifications. After the first answer from the neural network you often need to rephrase the question, add a detail or ask it to rewrite a fragment differently, and here short voice replies save time at every step of the dialogue, even if each is small on its own. Over a dozen such replies in one conversation a noticeable difference adds up, although none of them looks long by itself.

The third is conversational notes that later turn into a prompt: say aloud what you want to get from the neural network without worrying about structure, and then ask the same neural network to put the text in order. Here dictation wins precisely because there is no inner editor that slows typing: what is said aloud rarely comes out without stumbles the first time, but it is put into words faster than it is typed. How differently neural networks recognize and understand Russian speech at the input is a separate topic, covered in the article on the best neural networks for converting speech to text in Russian (in Russian).

The fourth scenario is found among those who work with code: dictating comments for a prompt, a bug description or a technical brief where Russian text is mixed with English terms and function names. Here accuracy matters more than speed: the same error in a variable name, repeated by recognition several times in a row, is more annoying than slow typing. A program with a personal dictionary remembers such words after the first correction, while without a dictionary you have to fix the same thing by hand every time.

What all four scenarios have in common is that the text must land exactly in the chat window of the neural network, not in a separate recording window that you would then have to copy from. A program with a hotkey such as Potok solves this directly: it inserts the recognized text where the cursor is, regardless of which window is open, and without an extra copying step between apps.

• Voice input in Cursor
• Voice input for ChatGPT: voice mode and text dictation
• Voice input in Claude: how to dictate text into the message field
• Dictation not working in a specific app on iPhone
• Wispr Flow: how much it costs and how to pay from Russia
• Voice input in ChatGPT: how the built-in dictation works
• Voice input in Claude: how to dictate text in Russian
• Voice input in Claude Code
• Voice input in Cursor: built-in dictation and Potok
• Diktuy: price and payment from Russia, how it works
• WhisperU: price and payment from Russia
• Wispr Flow: price and payment from Russia
• Restoring punctuation and capital letters: how voice input places marks without the word "zapyataya" (comma)
• In which countries 35 voice input services are registered

All articles in this section, in Russian (238): potok.now/golosovoy-vvod-dlya-neyrosetey

What to open first

• if you have never used voice input before, start with a comparison of voice input programs: starting from your platform and payment method, it is easier to choose one option than to try them all;
• if you already dictate ordinary text and want to move to prompts, look at the breakdowns for specific chats: voice input for ChatGPT and voice input in Claude; the switch takes a couple of minutes of getting used to it;
• if you are interested specifically in the quality of recognizing terms and mixed speech, start with the breakdown of how neural networks handle the Russian language with English inserts;
• if you write code and dictate technical prompts, look at voice input in Cursor: it covers example phrases and typical recognition errors for terms such as “zakommit” (commit), “pul-rekvest” (pull request) and “merdzh” (merge).

Who this section is not for

Voice input for neural networks is not a replacement for the neural network itself and not a guarantee of a good answer: it speeds up entering a request, not the quality of the request. Potok is a cloud service, recognition runs on servers in Russia, and it does not work at all without the internet; you will have to type by hand if there is no network. Potok also does not understand voice commands for editing a finished neural network answer: correcting the text after the answer still has to be done by hand or with the next prompt.

For those who type prompts in one or two short paragraphs and do it rarely, switching to dictation is unlikely to save noticeable time: the difference between voice and a keyboard becomes tangible on long and frequent texts, not on a single short question once a week.

Frequently asked questions

Related software: terms compared

Primary sources: speakflow.ru · speakflow.ru · speakflow.ru · blog.speakflow.ru