Most online transcription services work the same way: you upload your file to their servers, wait, and get text back. That's fine for a public video, but not for a client call, a doctor's voice note or an interview you promised to keep private.
There's another way. Modern browsers can run a speech recognition model on your own computer, so the audio never leaves it. This guide shows how to do that with Spokly's free audio and video to text tool, what to expect the first time, and how to get cleaner results.
How on-device transcription works
When you transcribe on your device, three things happen inside your browser:
- Your browser reads the audio from the file you chose. Video files work too; only the sound is used.
- A speech model runs on your computer. Spokly uses Whisper, an open-source speech recognition model, built to run in the browser.
- The text appears on the page, part by part, as the model works through the recording.
Nothing is sent to Spokly's servers at any point. You can check this yourself: once the model has downloaded, open your browser's developer tools, go to the Network tab and start a transcription. You'll see no upload of your file.
This is also why the tool can be free and unlimited. Your computer does the work, so there's no server bill for each minute of audio.
Step by step: transcribe a file for free
- Open the audio and video to text tool on the Spokly home page. No sign-up is needed.
- Drop your file onto the box, or click it to choose a file. MP4, MOV, MP3, M4A, WAV, OGG and WebM all work.
- Choose the spoken language. "Auto-detect" works, but picking the right language usually gives better results, especially for short clips.
- Choose Fast or Accurate (more on this below).
- Click Transcribe for free and keep the tab open while it works.
- When it's done, read through the text, fix anything that's wrong, then Copy it or Download TXT.
If your file is stored online, you can paste a link instead of uploading. Public links from Dropbox, Google Drive and OneDrive work, as do direct links to an audio or video file, up to 50 MB. In that case the file has to be fetched from where it's stored, but it's passed straight to your browser and never saved by Spokly.
The first time: a one-time model download
The first time you transcribe, your browser downloads the speech model. You'll see a progress message with the size. After that, the model stays in your browser's storage, so the next transcription starts right away.
How big is it? It depends on your browser. Browsers with WebGPU (a way for web pages to use your graphics card) use a version built for it, and those without use a version built for the processor. Spokly shows the exact size next to Fast and Accurate before you start, so you know what you're downloading:
| Choice | Without WebGPU | With WebGPU |
|---|---|---|
| Fast | about 125 MB | about 200 MB |
| Accurate | about 385 MB | about 565 MB |
A few things to know about this download:
- It happens once per browser, not once per file.
- If you clear your browser's site data, it will download again next time.
- On a slow or metered connection, start with Fast. It's the smaller download.
Fast vs Accurate: which one to choose
Both options use the same kind of model in two sizes.
Fast is the smaller model. It downloads quicker and transcribes quicker. For clear speech in a common language, such as a YouTube voice-over or a lecture recorded with a decent microphone, it's often all you need.
Accurate is the larger model. It's slower and the download is bigger, but it usually handles harder audio better: accents, dialects, fast talkers, background noise and less common languages.
A simple way to decide:
- Try Fast first on a short clip.
- If the result needs a lot of fixing, switch to Accurate and run it again.
- For Arabic dialects, start with Accurate. Our guide to transcribing Arabic audio explains why.
How long it takes depends on your computer and on the length of the recording. A recent laptop works much faster than an old one, and a computer with WebGPU is usually faster than one without. Let the tab stay open and in front while it works; some browsers slow down pages running in background tabs.
Limits to know about
- Length: up to 60 minutes per file. For a longer recording, split it into parts with any audio editor and transcribe each part.
- Links: up to 50 MB per file. Larger files can still be uploaded directly from your computer.
- Memory: very long files use more of your computer's memory. If your browser struggles, close other tabs or split the file.
- Speakers: the transcript is one block of text. It doesn't label who is speaking, so for interviews you'll add names yourself while editing.
Tips for cleaner audio (and a cleaner transcript)
The model can only write what it can hear. A few habits make a real difference:
Before you record
- Get close to the microphone. A phone held near your mouth often sounds clearer than a good microphone across the room.
- Pick a quiet room with soft furniture. Bare walls and hard floors add echo, and echo makes words blur together.
- Turn off fans and notifications. Steady background noise is easier to ignore than sudden sounds, but silence is best.
- Speak at a natural pace. You don't need to speak slowly, just finish your words.
Before you transcribe
- Use the original file, not a copy that was compressed again by a messaging app.
- Trim long silences or music at the start and end. Long music sections can produce stray words.
- Set the language yourself if the recording mixes languages or starts with music.
After you transcribe
- Read it once from top to bottom. Most mistakes are names, brand names and technical terms. Fix those first.
- Search for words that sound alike in your topic and check them.
- Add paragraph breaks where the speaker changes topic. It makes the text much easier to use later.
What to do with your transcript
A transcript is a starting point. Here are some common next steps:
- Make subtitles. The same recording can become an SRT or VTT file with the subtitle maker. See how to add subtitles to a YouTube video.
- Write an article. A podcast or talk can become a blog post with some editing: add headings, cut repetition and turn spoken sentences into written ones.
- Keep notes. Paste the text into your notes app so you can search your recordings later.
Spokly is also adding AI writing tools that turn a transcript into summaries, posts and articles. Those are coming soon; the transcription itself stays free.
FAQ
Is it really free, with no limit on the number of files?
Yes. Transcription runs on your own device, so there's no per-minute cost to pass on. The only limits are the 60 minutes per file and what your computer can handle.
Does my audio get uploaded anywhere?
No. When you upload a file from your computer, it's read and transcribed inside your browser. If you paste a link instead, the file is fetched from that link and passed straight to your browser, without being stored.
Why is the first transcription slow to start?
Your browser is downloading the speech model once. You'll see the progress and the size. Next time, it starts right away because the model is kept in your browser.
Can I transcribe on my phone?
It can work on a recent phone, but the model download is large and transcription uses a lot of memory and battery. A laptop or desktop computer gives a smoother experience, especially for long files.
Which browsers work best?
A recent version of Chrome or Edge on a computer is the safest choice. Other modern browsers can work too, but speed and support vary, especially for WebGPU.