The best free ways to transcribe audio in 2026

Run an open speech model yourself

Whisper, OpenAI's open speech recognition model, and the tools built around it, whisper.cpp and faster-whisper being the two most used, can run on your own computer for free. You download a model file once, feed it an audio file from the command line, and it writes out a transcript. There is no per-minute cost, no account, and no upload to anyone's server, because nothing leaves your machine.

The catch is that this is a technical task, not a product. You need to install the software, pick a model size, and understand that bigger models are more accurate but slower and need more memory. A modern laptop can run a mid-size model on an hour of audio in a reasonable time; a five-year-old machine with 8 GB of RAM will struggle or need the smallest, least accurate model. There is no support line if the install fails or the output looks wrong, only forums and documentation.

Once it is set up, though, it is the cheapest option there is for someone who transcribes often. No file size limit beyond your own disk, no daily cap, and no account tied to your usage. If you are comfortable with a terminal and expect to transcribe regularly, the time spent setting this up once pays for itself many times over.

This route wins for volume and for privacy in the strictest sense, since the audio never touches a network. It loses for anyone who wants to open a file and get text back in the next two minutes without installing anything first.

Free tiers of automated transcription services

A large category of hosted services, FastScribe among them, let you upload an audio file and get a transcript back through a web page, no software installed and no command line. Most offer some form of free usage: a limited number of minutes, a cap on file size, or a small daily allowance before they ask for an account or a card.

The appeal is speed of entry. You do not evaluate a model or manage an install; you drag a file onto a page and wait. The trade-off is that free tiers change often and vary by provider, one might cap file length, another caps total minutes per month, another requires a signup before you see the first result. Check the current limits on whatever service you are looking at rather than trusting a number you read somewhere else, because these move.

The other question worth asking before you upload anything is where the file goes and what happens to it afterward. Some services route audio through a third-party AI provider to run the model; others process it on their own servers. That distinction matters more for a sensitive recording than for a grocery list note to yourself, so it is worth a look at the provider's own description of its process before you commit a file you would rather not have sitting somewhere indefinitely.

This category wins for anyone who wants a browser, a file, and a transcript, with no setup and no ongoing technical maintenance. It loses if you transcribe constantly, since free tiers are built for occasional use, not as a replacement for the tools above.

Want to try it now? Upload a file to FastScribe. Your first one is free, no signup.

The tools already built into your phone, browser, and apps

Before installing or uploading anything, check what you already have. Most video hosting platforms generate automatic captions for uploaded video, and you can often copy that caption text out as a rough transcript at no extra step. If the audio you care about is already sitting inside a video you or someone else uploaded there, this is the fastest route of all: no new tool, no new upload.

Phone voice memo apps on modern phones commonly offer built-in transcription of a recording you have already made, turning a voice memo into text on the device without any other app. If your source audio started life as a phone recording, check that app first; it may already have done the work.

Word processors with dictation are a different tool for a related problem, and it is worth being precise about the difference. Dictation transcribes live speech as you talk into a microphone in real time; it does not take an existing audio file and transcribe it after the fact. If what you have is a recording someone else made, or a meeting you already captured, dictation will not help you, it needs a live voice, not a file. It is the right tool only when you are willing to replay your recording out loud into a microphone while the word processor listens, which works but is slower and less accurate than feeding the file to a proper transcription tool directly.

This category wins when the transcript you need is a byproduct of a tool you are already using for something else. It loses as a general-purpose solution: coverage is uneven across platforms and apps, and none of these were built as a dedicated transcription tool, so quality and features trail purpose-built options.

Doing it by hand

Typing what you hear, pausing, rewinding, typing again, remains a completely valid method. It costs nothing but time, and for a two-minute voicemail or a short quote you need to check, it is often genuinely the fastest option, faster than installing anything or waiting on an upload.

It does not scale. An hour of clear audio typically takes several hours to transcribe by hand well, more if the speaker is fast, the topic is technical, or there is more than one voice. For anything beyond a few minutes, the time cost adds up quickly, and most people underestimate it going in.

The one real advantage beyond cost is depth of familiarity: typing every word forces you to hear every word, which some researchers and writers value as a genuine first pass of analysis, not just a transcription step. If the recording matters enough that you would review it closely anyway, doing it by hand folds that review into the same pass.

This wins for very short audio, for maximum control, and for cases where the act of transcribing is itself part of the work. It loses badly on anything long or recurring, where the hours involved outweigh any tool's imperfections many times over.

How to actually choose between them

Start with volume. If you transcribe audio regularly, weekly or more, and are comfortable installing software, an open model on your own machine is the cheapest route over time and the most private, since files never leave your computer. If you transcribe occasionally, a hosted free tier gets you a result faster than any setup would justify.

Next, check what you already have. Before uploading a file anywhere, look at whether a video platform's captions or your phone's built-in transcription already covers it. Five seconds of checking can save an upload entirely.

Then weigh length against your patience. A very short clip is often fastest done by hand. A long file is where a model, whether self-hosted or a hosted free tier, earns its keep, since the time cost of typing scales with length in a way that machine transcription does not.

Finally, weigh privacy against convenience honestly. Running a model yourself keeps a file entirely on your device. A hosted service moves the file across a network to get processed, so understand where it goes and how long it is kept before you upload anything sensitive. None of these choices are wrong; they serve different situations.

Where FastScribe fits, and where it does not

FastScribe sits in the hosted free tier category above, and it is worth being specific about what that means in practice. Your first file transcribes free with no signup, up to 10 minutes and 50 MB, which is enough to judge the output on your own audio before deciding anything further. A free account raises that to 5 files a day, up to 30 minutes each, with your transcript history kept for 7 days.

Audio is processed on FastScribe's own servers and deleted the moment your transcript is ready, it is never sent to a third-party AI service. Exports come as TXT, SRT, and VTT on every tier. If you need more than the free tiers cover, longer files, DOCX export, batch uploads, or a history that does not expire, Pro is $12 a month and raises the per-file limit to 5 hours.

Where FastScribe is not the right choice: if you transcribe heavily and constantly, someone running dozens of hours a week, an open model on your own hardware will be cheaper over a long enough timeline, and it is worth the setup effort for that volume. And FastScribe does not identify who is speaking, it produces a transcript of the words, not labeled speakers, so if diarization, telling voices apart automatically, is what you actually need, look for a tool built specifically for that.

For the common case in between, a file you want turned into clean, private text quickly without installing anything, FastScribe is built for exactly that, and the first file costs nothing to try.

Where FastScribe fits

FastScribe fits the person who has an audio file right now and wants clean text back in minutes, without installing anything or sending the recording through a third-party AI service: audio is processed on our own servers and deleted the moment the transcript is ready. The first file is free with no signup, up to 10 minutes and 50 MB, and a free account extends that to 5 files a day, up to 30 minutes each, with 7-day history. Pro is $12 a month for files up to 5 hours, DOCX export, batch uploads, and history that is kept. It is not the right fit if you transcribe heavy volume constantly, an open model on your own hardware is cheaper at that scale, or if you need speaker diarization, telling different voices apart automatically, which FastScribe does not do.