Free transcription software in 2026: what works and what has a catch

"Free" means different things depending on the software. Some tools give you a permanently free tier with real limits. Some give you a trial that expires. Some are free but require technical setup. One is free only if you type everything yourself.

Here's what seven commonly recommended tools actually give you, tested in 2026.

OpenAI Whisper

Free forever, no account, no limits. The catch: you need to install it yourself.

Whisper is an open-source model released by OpenAI. You download it, run it via command line, and it transcribes audio files on your own hardware. There's no web interface, no file upload — just Python and a terminal.

On a Mac with Apple Silicon or a Windows machine with an Nvidia GPU, it runs fast. On a machine without a GPU, a 1-hour file takes 15-30 minutes instead of 4-5.

Accuracy: 95-97% on clean English audio. The large-v3 model is the most accurate; the base model is faster but less accurate.

The honest take: If you're comfortable with Python, this is the best free option. If you've never used a terminal, it's not worth the setup time.

TranscribTxt

Five files per month, up to 30 minutes each, no setup. Browser-based, works on any device.

Upload your file, get a transcript in 1-2 minutes, download as plain text. The Pro plan ($12/month) removes the monthly cap, adds SRT export, and increases the per-file limit to 600 minutes.

Accuracy is comparable to Whisper because it runs on the same underlying model. The free tier is enough for occasional use — a few podcast episodes, meeting notes, interview recordings per month.

The honest take: The easiest no-setup option for people who don't need high volume. The 5-file limit is the main restriction.

Google Docs voice typing

Free, built in, but it only works with live microphone input.

You open Google Docs, go to Tools → Voice typing, and speak. It transcribes what you say in real time. You cannot upload a recorded file and have it transcribed.

If you want to use this for a recording, you'd have to play the audio through your speakers while holding your microphone up to them — which degrades quality and doesn't work well in anything but a quiet room.

The honest take: Useful for dictation, not for transcribing existing recordings.

oTranscribe

Free manual transcription with audio controls built in.

You upload a file, it plays in your browser, and you type what you hear. It has keyboard shortcuts to pause, rewind 2 seconds, and control playback speed — the essentials for manual transcription. No account required, nothing stored on their servers.

Accuracy is 100% because you're doing the transcribing. Speed is whatever you can type. A 45-minute interview takes 3-4 hours.

The honest take: The right tool if you need exact transcription and automatic tools can't handle your audio. Particularly good for difficult accents, heavy background noise, or highly technical vocabulary.

Otter.ai

300 free minutes per month, 30 minutes per conversation. After that, it stops recording and you have to start a new session.

Otter's strength is speaker identification — it labels who said what, which AI tools without diarization don't do by default. It also integrates with Zoom and Google Meet to transcribe calls automatically.

The honest take: The free tier is limited for transcribing recorded files, but works reasonably well if you have a few short recordings per month. The speaker labeling feature is genuinely useful for interview and meeting audio.

Whisper.cpp

A reimplementation of Whisper in C++, which runs faster with less RAM than the Python version. Also free and open source.

Setup is slightly more involved than standard Whisper, but the speed improvement is significant on CPU — about 2-3x faster than the Python version without a GPU. Works well on older machines.

The honest take: For technical users who want Whisper speed without a GPU, this is worth the extra setup step.

Windows built-in dictation

Free, built into Windows 10 and 11. Works with microphone input only, not recorded files. Lower accuracy than Whisper-based tools on accented speech.

The honest take: Fine for quick notes while speaking. Not a transcription tool for recorded audio.

How to choose

Here's a simple decision tree:

Technical setup OK + high volume + no cost: Whisper or Whisper.cpp locally.

No setup, occasional use (under 5 files/month): TranscribTxt free tier.

Speaker identification matters: Otter.ai.

Audio is too messy for AI: oTranscribe for manual work.

Dictating while working, not transcribing a recording: Google Docs or Windows dictation.

The main difference between paid and free AI transcription is volume limits and format options (SRT export, timestamps), not accuracy. On clean audio, a free Whisper-based tool and a $30/month enterprise tool produce similar results.