How to Transcribe Audio on Windows

Built-in Windows features

Voice typing

Press Windows + H to dictate into any text field. It types what you say into the microphone; it doesn't read files.

Live Captions

Windows 11 can caption any audio playing on the PC in real time, shown in a bar on screen. It's an accessibility feature for following along, not a transcript you can save.

Both are useful, but for a 40-minute interview you want a document. Dictation vs transcription explains why live speech and recorded files are handled differently.

Microsoft 365: Transcribe in Word

Word for the web has had a Transcribe feature for Microsoft 365 subscribers: you upload an audio file, and Word produces a transcript with timestamps and speaker separation that you can insert into a document. It's convenient if your plan includes it, though it has had a monthly upload allowance and a limited set of languages. Check whether your account shows Transcribe under the Dictate menu in Word for the web.

Run a speech model locally

OpenAI's Whisper is open source and runs on Windows. With a recent NVIDIA graphics card, larger models run quickly; on a CPU alone, smaller models are practical and large ones are slow. Options range from the original Python package (which needs ffmpeg) to whisper.cpp and graphical apps built on it. How to run Whisper locally covers setup and model sizes. Local transcription keeps files on your PC and costs nothing per file, at the price of setup time.

Upload to a service: mydubly

1. Open audio to text or video to text.
2. Add the file. Videos stay on your PC; only the audio is extracted and sent.
3. Choose the spoken language, or one of 21 languages for a translated transcript.
4. Download plain and timestamped transcripts and SRT or VTT subtitles.

Pricing is per minute of audio: 1 credit, a tenth of a cent, and new accounts start with 100 free credits. Files can run up to 2 hours.

Getting audio out of Windows apps

Sound Recorder

Saves recordings as M4A by default, with other formats available in its settings. Find files in the app's recordings folder or with Show in folder.

Clipchamp and other editors

Export the video, or an audio-only file where offered.

Teams, Zoom and Meet

Download the meeting recording; see transcribing a Teams recording.

Phone recordings

Copy them over USB or from cloud storage; phone call recording transcription covers call recordings specifically.

For a video you only need the words from, the free extract audio from video tool makes a small audio file in your browser without uploading.

Choosing between them

Following live audio as it plays

Live Captions.

Occasional files, Microsoft 365 plan

Word's Transcribe, if available.

Privacy first, capable PC

Local Whisper.

Speed, translation, subtitle files

A hosted service.

Limits and trade-offs

• Local models without a capable GPU can be slower than the recording itself.
• Hosted services send audio off your PC; with mydubly it's deleted with the results within 30 minutes of delivery. Cloud speech processing privacy lists questions for any service.
• mydubly transcripts don't label speakers.
• Poor recordings limit every method; how to check audio quality covers a quick pre-check.

Videos, not just audio

Everything above applies to video files too. Screen recordings made with the Snipping Tool or the Xbox Game Bar, camera clips and downloaded meeting recordings can all be transcribed, and for videos the transcript usually comes with subtitle files you can add back to the video. If you record screens for tutorials, check that the microphone was captured as well as system sound; screen recording audio settings covers the common mistakes, such as recording only the computer's sound. The how to transcribe a video guide walks through the whole process for video files.

Comparing options on your own recordings

Accuracy depends more on the recording than on the tool, so test before choosing for a large batch. Take a 5-minute excerpt with the hardest audio you have (a noisy room, an accent, technical terms) and run it through two options. Compare names, numbers and the noisiest minute against the recording. Speech recognition and accents explains why results vary between speakers, which is worth knowing before you judge a tool on one person's voice. Keep the excerpt and your corrected version; they make a handy reference set whenever you try a new tool later.