Audio to Text Free: How Big a File You Can Upload Without an Account

The ladder

Size and length move together on this ladder: 1 MB is about 5.9 seconds of audio, and 100 MB is about 9 minutes 54 seconds.

• 524,338 bytes (3.0 s): 200 in 1.57 s - 12 words.
• 1,048,598 bytes (5.9 s): 200 in 20.93 s - 346 words.
• 2,097,298 bytes (11.9 s): 200 in 13.84 s - 38 words.
• 5,243,038 bytes (29.7 s): 200 in 7.09 s - 91 words.
• 10,485,822 bytes (59.4 s): 200 in 12.74 s - 101 words.
• 12,583,042 bytes (71.3 s): 200 in 38.99 s.
• 15,728,786 bytes (89.2 s): 200 in 15.95 s.
• 16,100,000 bytes (91.3 s): 200 in 16.91 s.

Where it stops

The ceiling sits between 16,100,000 and 16,776,000 bytes. We tested both sides of it more than once. Files of 15.9 MB and 16.1 MB both pass, and 16,776,000, 16,777,000 and 16,777,306 all come back 503 with Cloudflare error 1102 after about six seconds.

• 16,100,000 bytes: 200.
• 16,776,000 bytes: 503, error 1102.
• 18,874,526 bytes: 503, error 1102.
• 52,428,806 bytes: 503, error 1102.

It's the size, not the length

That's worth checking, because a 20 MB file is also a two-minute file, and it's fair to ask which one breaks it. We rebuilt 20 MB at 96 kHz so it ran 54.6 seconds instead of 118.9. It still failed.

Then a 16.1 MB file at 91.3 seconds still passed. Short and fat fails, long and thin passes, so what you're hitting is the size of the request, not the clock. At 100 MB the failure changes shape - Cloudflare returns 413 Payload Too Large in 0.16 seconds, before any processing. Nothing was transcribed and nothing was stored, because nothing arrived.

How long the upload takes

Wall-clock times moved around a lot on the same ladder: 1.57 seconds for the smallest file, 38.99 seconds for the 12 MB one, 7.09 seconds for the 5 MB one. Bigger doesn't mean slower in any clean way.

Two runs of the 1 MB file both took about 21 seconds while the 5 MB file took 7. Don't budget your day around a single measurement, and don't read a slow upload as a sign the file is too big.

One file came back with 346 words from six seconds of audio

This is the strangest result on the ladder, and we ran it twice to be sure. The 1 MB file - 5.9 seconds of one sentence - returned 1,621 characters and 346 words, most of them "at 8 o'clock in the morning" over and over. The next rung up, 11.9 seconds, returned a normal 38 words.

Loop a short clip and the model can get stuck repeating a phrase instead of stopping. A transcript much longer than the audio can hold is the fingerprint. Check the word count against the length before you trust the text.

What 16 MB means in minutes of audio

Raw stereo WAV is the worst case, and it's what we used. Nobody uploads a podcast that way. Here's the same ceiling in formats people actually use.

• 16 kHz mono WAV: 32,000 bytes per second - about 8 minutes of audio.
• MP3 at 128 kbps: 16,000 bytes per second - about 17 minutes.
• MP3 at 64 kbps mono: 8,000 bytes per second - about 33 minutes.

What to do when your file is too big

Re-encode rather than chop. The 16 MB wall is on the request, and speech compresses hard. The same audio as mono 16 kHz MP3 runs roughly 2,000 times smaller per second than the stereo 44.1 kHz WAV we used here.

• Convert to MP3 or AAC at 128 kbps or lower before uploading.
• Drop to mono. Two channels of speech carry the same words twice.
• Split a long recording into pieces only if converting still leaves it over the line.

What free means here

No account, no key, no queue, and no dashboard to check usage, because there's nothing tying uploads to a person. Every file in this test went to the public endpoint with no sign-in.

The limit we hit is per request, not per day. This run pushed more than twenty files through in a single session without being stopped.

What happens to your file here

Your file does leave your browser. It travels over an encrypted POST to /api/transcribe, is held in memory for one pass, and is dropped the moment the response is written. No copy is kept, nothing is used for training, and there's no account to attach it to.

Why we built the ladder out of WAV

Because it's the worst case. Uncompressed stereo audio hits the wall sooner than anything you'd actually upload, so every number here is a floor rather than a ceiling. If your file is an MP3, you have more room than this table shows.

Limits of this test

We didn't test MP3 or M4A at the top of the range, and timing varied too much run to run to quote as typical.