Audio-to-text transcription: comparing services and when you need voice input, not a transcriber

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

Transcription turns an already recorded audio or video file into text as a whole, while voice input turns speech into text at once, while you speak, right in the field with the cursor. For lectures, interviews and calls you need a transcriber. For quick dictation of emails and notes, Potok (potok.now) fits, with the Ctrl+Space hotkey and 14 days free.

Transcription and voice input are different tasks

The word “transcription” and the phrase “voice input” are often confused in search, although in meaning they are different tools for different situations. Transcription is work with an already existing recording: you have a file on disk (an interview, a lecture, a call, a podcast, a voice message), and the task is to get a coherent text out of it as a whole, usually split into paragraphs or with timecodes, sometimes with an indication of which speaker said what. The recording itself has already happened in the past, and the service processes it as a finished piece of audio or video fed in as input.

Voice input works differently: the text appears not after a file is processed but at the moment of speech. You turn the tool on, as a rule by holding a key, say a phrase and at once see the result where you would type by hand: in an email, in a messenger, in a search field, in a code editor. No intermediate file appears, processing goes in short phrases almost in real time, and the result at once becomes part of the document, not a separate export.

The difference also shows in what it makes sense to ask of a service. For a transcriber, questions about the length of the uploaded file, support for mp3/wav/mp4 formats, splitting by speaker and export to srt or docx with timecodes are relevant. For a voice input tool these questions make no sense: what matters there is the response speed for a short phrase, the stability of the hotkey and how accurately punctuation is placed right during dictation.

The confusion is increased by the fact that both classes of tools solve what looks like a similar “speech to text” task and often end up in the same list in search. But the difference is easy to check with one question: do you have a ready file that needs to be transcribed, or do you want to dictate text right now, with no recording and no upload. The first case is about transcription, the second about voice input, and this question is worth starting from when choosing a tool, not the name of the category on the service's website.

How to evaluate a transcription service

If the task is precisely to transcribe a finished recording, when choosing a service it is worth looking at several practical parameters, and not only at the general impression of the website. First, which file formats are accepted and whether there is a limit on length or size: some services cut large interviews into pieces or limit the free plan to a few minutes. Second, how readable the result is in Russian: if the recording contains professional terms, names, a mix of Russian and English, the quality of recognition and punctuation placement may drop noticeably.

Third, processing speed matters: transcription is almost always batch, you upload a file and wait until the service returns the finished text, and this time is worth building into your work plan, especially for long recordings. Fourth, whether you need diarization, the labeling of which of several speakers said what: for interviews and meetings this is often a key function, not a pleasant addition. Fifth, where the recording itself goes: some services keep uploaded files on their servers for some time, and for confidential calls this is worth clarifying separately, before uploading and not after.

Finally, compare the payment model. Some services charge per minute of processed audio, others sell a subscription with a limit of hours per month. For a one-off interview a one-time payment is better, for regular transcription of calls a subscription with a clear limit. Separately check the output format: some need a simple txt with no markup, others srt subtitles with timecodes for video editing, still others docx right away with headings and paragraphs for publication. Not all services hand over all three formats equally easily, and converting from one to another by hand takes the time that transcription was supposed to save.

Transcription services: briefly by category

The transcription market is organized into at least three categories of solutions that differ in format, and each has its own strengths and limits, regardless of specific names.

Cloud APIs for developers. These are services that do not give the user a ready interface but hand out speech recognition as a line in your own code: you send an audio file to the server on request and get the text back in a structured form. This option suits you if transcription is needed not by a person by hand but as part of someone else's product, for example automatic transcription of calls in a CRM. The entry threshold is higher: you need at least a little code and an understanding of how to connect an API, but integration into an existing process turns out more flexible.

Desktop apps. The program is installed on a computer, the file is chosen through the ordinary Open window, and processing goes locally or through the developer's server, depending on the specific app. Some such programs can work fully without an internet connection, which is critical for recordings that must not be uploaded outside your own computer. The flip side is that processing usually loads the computer's own processor and memory, and on long recordings this is noticeable in speed and in fan noise.

Online services in a browser. The lowest entry threshold: you go to a website, upload a file through a form, wait and download the finished text, with nothing to install. This format is convenient for a one-off task, but it means the recording goes to someone else's server as a whole, and it is worth checking the file storage and deletion policy in advance, especially if sensitive data is heard in the recording.

When you need on-the-fly voice input instead of transcription

Sometimes behind the query “audio-to-text transcription” there is actually another, simpler need: not to transcribe an already recorded conversation but to stop typing by hand right now. If you are looking for a way to dictate an email, a code comment, a note in a notebook or a reply in a messenger, you do not yet have a file to upload, and you basically do not need one. Here a voice input tool takes the place of a transcriber: it does not transcribe finished recordings but turns your live speech into text at the moment of talking, with no intermediate file and no waiting for processing.

This is how Potok is built. To be honest: it is a dictation tool, not a service for transcribing files. You hold the Ctrl+Space hotkey, say a phrase, release, and the finished text with punctuation is at once put into the field where the cursor was: in email, a messenger, a document or a line of code. In Potok you cannot upload a finished audio recording of an interview and get a transcript from it; for that you need a separate transcription service from the categories above. But if the task is to speak and see the text at once, without being distracted by the keyboard, a transcriber is excessive here: it requires you first to record a file, then upload it somewhere and wait for processing, while voice input removes this path entirely.

Potok works on Windows 10, Windows 11 and macOS; for the first 14 days access is full and free with no card, after that comes Free at 2500 words per week or Pro at 690 ₽ per month (499 ₽ per month when paying for a year). Recognition runs on its own model trained for Russian speech, including terms and a mix of Russian and English.

Potok is a cloud service, recognition runs 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