How to work through a folder of recordings without uploading any of them
There's no batch transcription button: the tool takes one file at a time. Here's how to get through a folder of recordings in one sitting, uploading nothing.
There is no batch transcription in FreeTranscribe. It takes one file at a time: you pick a recording, it runs, you export the text, you pick the next one. No folder drop, no queue, no overnight job list. What makes a stack of twenty recordings practical anyway is that the speech model downloads once and is then cached, so every file after the first starts straight away. The rest is a naming habit, one long sitting, and an index file at the end. Nothing you transcribe leaves your computer.
Batch transcription in a browser still means one file at a time
Worth being blunt about what you are signing up for. You will be at the machine, picking files. For a term of lectures or a season of interviews, that is twenty or so trips to the file picker.
That sounds worse than it is. The picking takes seconds, the waiting is where the time goes, and the waiting does not need you. What you are planning is one long run with check-ins, not twenty separate tasks.
The reason to put up with it is the same reason the folder exists. If those are interviews you promised not to share, or calls with a client, then “nothing uploaded” is the whole point and the file picker is the price.
The model downloads once, so only the first file is slow
The first run pulls down OpenAI’s open-source Whisper model, about 200 MB, and the browser caches it. That download is the part people remember as slow, and it happens once.
So the shape of a long sitting is: first file, wait for the model, then transcription. Second file onwards, transcription only, starting as soon as you pick it. Keep the tab open across the whole run and you pay that download once for the entire folder.
Name the folder before you start, not after
Do this first, because fixing it afterwards means opening files to work out what they are.
Put the whole set in one folder, one per term, project or season. Rename every recording to the same pattern, starting with an ISO date so sorting by name also sorts by time:
2026-03-04 kickoff shaw.m4a2026-03-11 site visit shaw.m4a2026-03-18 review shaw.m4a
Date, then two or three words you would actually search for. Avoid interview1, interview2. If the recorder gave you REC0042.WAV, renaming twenty of those takes about five minutes.
Check the formats while you are in there. MP3, WAV, M4A, AAC, FLAC, OGG, MP4, MOV, WEBM, MKV and AVI all work, and for video only the audio track is read. Anything else in the folder is better found now than halfway through the run.
Keep the tab open and the machine awake
The run lives in the tab. Close the tab and the model cache survives, but the run does not.
Plug the laptop in. Then go into your power or battery settings and find the setting that puts the computer to sleep, as distinct from the one that only turns the screen off, and set it to never for the duration. The screen going dark is fine. The machine sleeping is not, and on a laptop that usually includes closing the lid.
Leave the graphics card alone while it works. The recognition runs on your GPU through WebGPU, so a game or a video export at the same time is competing for the same hardware. Browsing and writing are fine.
Export each transcript as soon as that file finishes, before you pick the next one. It takes seconds, and it means an interruption costs you one file rather than the afternoon.
How long a stack of recordings takes
In our test, a desktop with a graphics card ran at about 1.5 times real time, so an hour of audio comes back in roughly 40 minutes. A thin laptop takes roughly as long as the recording itself. Both are our own measurements, not a promise about your hardware.
The table assumes recordings of about an hour each. Halve the audio and halve the time.
| Recordings, about an hour each | Rough run time | What to do while it runs |
|---|---|---|
| 3 | About 2 hours on a desktop, about 3 on a thin laptop | One evening at the desk. Export each one as it finishes. |
| 5 | About 3.5 hours on a desktop, about 5 on a thin laptop | Half a working day. The gaps are long enough for other work. |
| 10 | About 7 hours on a desktop, about 10 on a thin laptop | A full day, or two evenings. Split it rather than babysitting past bedtime. |
| 20 | About 13 hours on a desktop, about 20 on a thin laptop | Two or three sittings, with sleep turned off and a check-in every couple of hours. |
Twenty one-hour recordings is the case worth thinking about before you start. On a thin laptop that is most of a day of machine time with a person nearby to click. If that does not fit your week, skip to the last section.
Export the same way every time, and build one index
Pick the export format at the start and stick to it. TXT if you want to read and search the words, SRT if you need to jump back to a moment in the audio, since every cue carries a start time. SRT vs VTT vs plain text covers which one a given job wants.
Give each transcript the same base name as its recording, so 2026-03-04 kickoff shaw.m4a becomes 2026-03-04 kickoff shaw.txt. Sorted by name, the audio and its text sit next to each other.
Then write the index. One plain text file in the folder, index.txt, with a line per recording:
2026-03-04 kickoff shaw 62 min scope, budget, march deadline
Date, topic, length, three or four keywords. It takes a minute per file and gives you a table of contents you can read in one screen. Because it is plain text sitting with the transcripts, your operating system’s own search may index the whole folder, so one search box answers both “which recording was that in” and “what was said”. Many text editors have a find-in-files command that will search the folder either way. Try it on the first two files, so you know it works before you have twenty.
Read the transcripts through before relying on them. The base model is weakest on names, technical terms, strong accents and noisy rooms, and across twenty files the same misheard surname turns up twenty times, which at least makes it quick to fix with find and replace.
When a queue service or a command-line tool is the better answer
For a few dozen files, a sitting with the tab open is fine. For several hundred, it is not, and there is no version of this post that makes it fine.
Two honest alternatives. Paid transcription services generally take a whole folder, queue it and work through it on their own servers, so you get a notification instead of a day at the desk. The trade is that the recordings are uploaded and you pay for the minutes. When that trade is acceptable for your material, it is the right tool, and our cost comparison of transcription services has the published prices.
The other is running a speech model yourself from a command line on your own machine. You install a runtime, point a loop at the folder and walk away. Nothing is uploaded and nothing needs a browser tab, so it can run overnight without you. The trade is setup: a terminal, an install and some reading before the first file goes through. For one folder, once, that is a poor deal. For a folder that refills every month, it is usually the right one.
Neither is what we are. We are the free, no-account option that handles one file well and asks you to come back for the next one. For a term, a project or a season, that is often enough. To start on the first recording in your folder, open it in your browser. It runs on desktop Chrome or Edge, it is English only for now, and the audio stays on your machine.
Frequently asked questions
Can I drop a whole folder onto the page? No. It takes one file at a time and there is no queue behind the scenes.
Does the model download again for every file? No. It downloads once, about 200 MB, and the browser caches it. Keep the tab open across a sitting and every file after the first starts straight away.
Can I leave a long run going unattended? A file keeps going as long as the machine is awake and the tab is open, so turn off sleep and stay plugged in. Someone is still needed between files to pick the next one, so plan for check-ins rather than a true overnight job, and treat an interrupted file as one to run again.
Is there a limit on how many files I can do in a day? No. There is no account, no length cap and no per-minute charge, so the limit is how long you are willing to sit there. Why free transcription sites cap you at 30 minutes explains why that is unusual.
What if some of the recordings are not in English? The tool is English only for now, so those files are not covered yet.
Sources, checked 15 September 2026
- None fetched. This post makes no claim about any named third-party product, platform or standard, so no external sources were needed. The speed figures are from our own test of the tool, and the format list and export options are its current behaviour.