How big is an hour of audio?

An hour of audio has no single size, because size depends on how the audio was encoded. At the light end, a one-hour recording saved as a 128 kbps MP3 comes to about 58 MB. At the heavy end, the same hour stored as an uncompressed CD-quality WAV file – 44.1 kHz, 16-bit, stereo – is roughly 600 MB. That is more than a tenfold spread for the exact same sixty minutes.

Between those two poles sits everything else. A 320 kbps MP3, the highest common MP3 setting, runs near 144 MB for the hour. A FLAC file, which keeps the WAV's audio perfectly intact, lands somewhere around 300 to 420 MB. The number you get is a direct read-out of one choice: the bitrate.

File size is not mysterious. For any constant-bitrate audio, the size in bits is simply the bitrate multiplied by the duration in seconds; divide by eight to get bytes. A 128 kbps stream over an hour is 128,000 bits per second times 3,600 seconds, which works out to about 58 MB. Double the bitrate and you double the file.

Uncompressed audio follows a slightly longer version of the same idea: sample rate x bit depth x channels x duration. CD-quality audio under the Red Book standard streams at 1,411.2 kbit/s – that is 2 channels times 44,100 samples per second times 16 bits – or 176,400 bytes every second. Everything below is just this formula, run for each format.

Uncompressed PCM audio, which is what a .wav file usually holds, stores every sample in full. At CD quality that is about 10.1 MB per minute. Five minutes comes to roughly 52.9 MB (about 423 million bits); a full hour reaches somewhere around 600 MB, give or take depending on whether you count in decimal or binary megabytes.

You can shrink that without compressing by capturing less. Speech-capture setups often drop to mono, 16 kHz, 16-bit – a data rate of 256 kbit/s, or about 1.92 MB per minute. That is roughly 115 MB for an hour: far lighter than CD-quality stereo, but still heavy. WAV is the format you reach for when you need the raw signal, not the one you keep for long-term storage.

MP3 sets its size entirely by the bitrate you chose. At 128 kbps – the common "good enough for speech and casual music" setting – an hour is about 0.96 MB per minute, or roughly 58 MB. At 192 kbps it climbs to about 1.44 MB per minute, near 86 MB an hour. At 320 kbps, the ceiling for standard MP3, it is about 2.4 MB per minute: Colin Crawley's calculator puts 5 minutes 17 seconds at 320 kbps at 12.68 MB, which scales to roughly 144 MB for the hour.

So the practical MP3 range for an hour of audio is about 58 to 144 MB – a spread of nearly 2.5x driven by nothing but the bitrate slider. If you are transcribing a podcast or any spoken-word file, 128 kbps is usually plenty, and it keeps the file at the small end. Converting a heavier recording down to a 128 kbps MP3 before you turn the audio to text is often the quickest way to get under a size limit.

AAC is the codec inside most .m4a files, and at the same bitrate it produces a file the same size as MP3 – a 128 kbps AAC hour is also about 58 MB. The difference is quality per bit. AAC generally sounds better than MP3 at a given bitrate, and 128 kbps variable-bitrate AAC is good enough to meet the ITU's bar for transparent stereo. If two files are the same size, the AAC one usually carries the better-sounding hour.

FLAC is the interesting middle. It compresses audio losslessly – the decoded file is bit-for-bit identical to the original WAV – yet it typically reduces the data to between 50 and 70 percent of its original size. An hour of CD-quality WAV, around 600 MB, becomes roughly 300 to 420 MB as FLAC with nothing lost.

This is why audio-specific compression matters. A general-purpose compressor like Zip barely dents a WAV, because it cannot model the shape of sound; FLAC reduces the same file to 50 to 70 percent of its size because it is built for audio. The catch is that FLAC is still far larger than an MP3, so it is a storage-and-archive format, not an upload-anywhere one.

All of these numbers collide with one specific limit the moment you try to transcribe. OpenAI's speech-to-text API – the same engine behind ChatGPT's audio features – caps uploads at 25 MB, and accepts only these input types: mp3, mp4, mpeg, mpga, m4a, wav, and webm (Last verified: 2026-07-13). Anything bigger has to be compressed or split into chunks under the cap.

Line the cap up against the math and the problem is obvious. 25 MB holds only about 2.5 minutes of CD-quality WAV, or roughly 25 minutes of a 128 kbps MP3. A full-hour 128 kbps MP3, at about 58 MB, is more than twice the cap; a 320 kbps hour, near 144 MB, is close to six times it; and a WAV hour blows past 25 MB about twenty-four times over.

This is why a one-hour recording so often bounces when you drop it into a chat window. If you want the full list in one place, we collected the current AI transcription limits, and there are practical routes around it in our guides on transcribing audio over 25 MB and whether ChatGPT can transcribe audio.

To size any hour of your own, start from the format. Uncompressed WAV: about 600 MB, or about 115 MB if it is mono 16 kHz speech. FLAC: 300 to 420 MB. MP3 or AAC at 320 kbps: about 144 MB. At 128 kbps: about 58 MB. The lower the bitrate, the smaller the file – and the sooner it fits under a service's limit. Longer files are their own topic; our guide to transcribing long audio covers the duration side.

The wall only matters if you are the one doing the splitting. Pepys handles long files on the server: its pipeline downloads the file, converts it to 16 kHz mono, then chunks it before transcribing, so an hour-long MP3 or a 600 MB WAV uploads whole with no manual cutting. It is pay-once rather than a subscription, which suits the bursty way most people record. If your hour is sitting past the 25 MB line, that is the gap it is built to fill – not a different answer to the size question, just a way to stop the size from being your problem.