Speech to text · On your PC · 100 languages

Free offline transcription for Windows

Turn recordings and meetings into accurate, editable transcripts on your own PC. Nothing is uploaded, there is no account, and it works without the internet.

Coming soon to the Microsoft Store · Windows 10 and 11

  • No upload
  • No account
  • No tracking
  • Works offline

At a glance

Everything happens on your PC

Transcribe brings the speech recognition to your recordings instead of sending your recordings to the speech recognition. That one decision changes what it costs, who can see your audio, and whether you need the internet at all.

0
Recordings uploaded. Speech recognition runs on your PC
No account, no telemetry
100
Languages, detected automatically or chosen by you
17×
Faster than real time with Standard on a laptop's integrated graphics
One hour in about 3.5 min
1 GB
For the app and both speech models. Nothing downloads later
$0
To start. Standard is free without limits
Pro is optional

How it works

From recording to clean text in three steps

  1. Drop in a file, or record

    Drag audio or video onto the window, or choose files: MP3, M4A, WAV, WMA, FLAC, MP4 and more. With Pro, record a meeting or a call straight from your microphone and your PC's own sound.

  2. Choose a model and transcribe

    Standard is fast and free without limits. Accurate is slower and better with noise, accents and names. Pick the language or let the app detect it, say how many people speak, and carry on working: long jobs run in the background.

  3. Check, fix and export

    The transcript is tied to the audio. Click any line to hear it, fix it in place, replace a misheard name everywhere at once, then copy the text or export plain text and SRT or VTT subtitles. Pro adds Word and PDF.

The import dialog in Transcribe: the Standard and Accurate models, the language, and how many people speak.
Drop in audio or video. Choose Standard for speed or Accurate for noise, accents and names, in 100 languages, and how many people speak.

Private by design

Your audio never leaves your PC

Cloud transcription services upload your recording to their servers. Transcribe runs the speech recognition on your own machine, so there is nothing to upload, no account to create, and no server that could keep a copy.

Your PC

Your recordingA file you import, or a meeting you record
Speech recognitionWhisper in whisper.cpp, on the graphics card or the processor
Your transcriptIn the library folder, with your edits and notes

Recordings, transcripts, notes and settings never leave this box.

The internet

  • Microsoft Store: checks your plan and shows prices, through Windows' own Store service. No data of the app's own is sent.
  • No server that receives your audio
  • No account, no sign-in
  • No analytics, telemetry, crash reports or ads

The app makes no connections of its own. It asks the Microsoft Store, through Windows' own Store service, whether you have Pro and what it costs, and that is all. The privacy policy lists every case, in plain words.

Features

Made for checking and correcting, not just for transcribing

Every automatic transcript is a draft. Transcribe is built around the part that takes your time: hearing a line again, fixing it, and moving on.

Tied to the audio

The line being played is highlighted and stays in view. Click a line or its time to play from there, at 0.75x to 2x.

Edit in place

Fix any line where it is. It saves itself, and undo and redo work across the whole transcript.

Find and replace

Fix a misheard name everywhere at once, then carry on reading.

Record meetings and calls Pro

From your microphone and your PC's own sound. The text appears a few seconds after people speak.

Who said what Pro

Speaker detection labels every line. Name the speakers once and every export follows.

Folders and notes

Folders and sub-folders keep clients, courses and projects apart. Notes sit beside each transcript.

Long recordings

Several hours work too. Jobs run in the background and continue after a restart.

Clean exports

Copy, plain text, SRT and VTT subtitles, with speaker names. Pro adds Word and PDF.

Search everything Pro

Search every transcript at once and jump to the moment it was said.

A look inside

Calm, fast, and made for long reading

The transcript reads like a page, the player sits below it, and your notes stay beside it. Light and dark, like Windows.

A transcript in Transcribe: timestamps on the left, the line being played highlighted in yellow, notes and speakers on the right, and the player with its waveform at the bottom.
An accurate transcript beside the audio, with your notes. The line being played stays highlighted; click any line to play it or fix it.
With Pro, record a meeting or a call from your microphone and your PC's sound. The text appears as people speak; a small recorder stays on top.
With Pro, record a meeting or a call from your microphone and your PC's sound. The text appears as people speak; a small recorder stays on top.
Name the speakers once and every line follows: in the editor, in text and subtitles, and in Word and PDF.
Name the speakers once and every line follows: in the editor, in text and subtitles, and in Word and PDF.

Speed and accuracy

Two models, measured on an ordinary laptop

Both models are OpenAI's Whisper, running in whisper.cpp on your PC. We timed them and counted their mistakes, and publish every run.

Measured on November 8, 2026 on a laptop with AMD Radeon integrated graphics (Vulkan) and 8 processor threads. Word error rate on 6 scripted recordings with synthetic voices, clean audio: real recordings with crosstalk and weak microphones make more mistakes. The method and every run.
ModelOn integrated graphicsOn the processorWord error ratePlan
Standard (Whisper small)17× real time: one hour in about 3.5 min3×: one hour in about 20 min2.0%Free, without limits
Accurate (Whisper large-v3-turbo)2.5×: one hour in about 24 min1×: one hour in about an hour1.1%15 minutes free, unlimited on Pro

Plans

Free for everyday transcription. Pro when you record meetings.

Free

$0forever

Everyday transcription with the Standard model, without limits and without an account.

Download for Windows
  • The Standard model without limits
  • 15 minutes of the Accurate model
  • 100 languages, detected automatically or chosen by you
  • The editor tied to the audio, with find and replace
  • Name speakers and set each line's speaker by hand
  • Folders, sub-folders and notes
  • Long recordings in the background
  • Copy, plain text, SRT and VTT export

Pro Yearly 48% less

$49.99a year

About $4.17 a month, billed once a year by the Microsoft Store.

Upgrade inside the app, from any Get Pro link or Settings › Plan.

Everything in Free, and:

  • Record meetings and calls with a live transcript
  • Speaker detection: who said what, on your PC
  • The Accurate model without limits
  • Word (DOCX) and PDF export with a title block
  • Search across every transcript and jump to the moment
  • A queue for many files
  • Custom words for names and terms

Pro Monthly

$7.99a month

The same Pro, month by month. Cancel any time in your Microsoft account.

Upgrade inside the app, from any Get Pro link or Settings › Plan.

Everything in Free, and:

  • Record meetings and calls with a live transcript
  • Speaker detection: who said what, on your PC
  • The Accurate model without limits
  • Word (DOCX) and PDF export with a title block
  • Search across every transcript and jump to the moment
  • A queue for many files
  • Custom words for names and terms

Prices are in US dollars; the Microsoft Store shows the price in your currency before you buy. Subscriptions renew until you cancel them in your Microsoft account (account.microsoft.com/services). If a plan ends, nothing is deleted: transcripts and exports stay, and the Pro features lock until Pro is active again.

Offline or cloud

Why offline wins for private recordings

Cloud services are quick and convenient. For recordings you are responsible for, the trade-off looks different.

Transcribe - OfflineA typical cloud service
Where your audio goesNowhere. It stays on your PCUploaded to the provider's servers
AccountNoneRequired
Internet connectionNot neededNeeded for every file
PriceFree, or a flat Pro subscriptionUsually per minute, or a monthly allowance
Who else can access itOnly people with access to your PCDepends on the provider's policies and staff
SpeedDepends on your PC: see the benchmarksFast, if your upload is

Questions

Questions people ask

Is Transcribe - Offline really free?

Yes. The Standard model has no limits, and Free includes 15 minutes of the Accurate model, the editor, folders, notes, and export to text and subtitles, with no account and no time limit. Pro is an optional subscription for meeting recording, speaker detection, the Accurate model without limits, Word and PDF export, search across every transcript, a queue and custom words.

Does it work without an internet connection?

Yes. The speech models are installed with the app, so transcription never needs the internet. The app only contacts the Microsoft Store to check your plan and show prices, and it remembers the Store's last answer so Pro keeps working while you are offline.

Is my audio uploaded anywhere?

No. Speech recognition runs on your PC with OpenAI's Whisper models in whisper.cpp. There is no server that receives audio, no account, no telemetry and no analytics. Exports are saved only where you choose.

How accurate is it?

Transcripts are made by Whisper, one of the most accurate speech recognition models available. On our test set of clean, scripted recordings, Standard got 2.0% of words wrong and Accurate 1.1%. Real recordings with noise, accents and people talking over each other make more mistakes, and like all automatic transcription it can occasionally write words that were not said. Check important passages against the audio: the editor makes that quick. See the benchmarks.

Which languages does it support?

One hundred languages, detected automatically or chosen by you. Accuracy varies by language and is best in widely spoken ones.

Can it transcribe Zoom, Microsoft Teams or Google Meet calls?

Yes, with Pro. Transcribe records your microphone and your PC's own sound, so it works with any call app, and the text appears while people speak. No bot joins the call. Tell everyone that you are recording. How to transcribe online calls.

Can it tell who is speaking?

Yes. With Pro, speaker detection labels every line with its speaker, on your PC. On Free you can name the speakers and set each line's speaker by hand. Either way, names appear in every export.

What do I need to run it?

Windows 10 version 1809 or later, or Windows 11, 64-bit, and about 1 GB of disk space. 8 GB of memory is recommended for the Accurate model. A graphics card with a Vulkan driver makes transcription much faster but is not required.

Is there a Mac, Linux or phone version?

No. Transcribe - Offline is a Windows app, made for Windows 10 and 11.

How do I cancel Pro?

In your Microsoft account, at account.microsoft.com/services, like any Microsoft Store subscription. If a plan ends, nothing is deleted: your transcripts and exports stay, and the Pro features lock until Pro is active again.

Your recordings stay on your PC.

Free to download. Transcribes without an account, without uploads and without the internet.

Download for Windows

Coming soon to the Microsoft Store · Windows 10 and 11