Getting an audio file under an attachment limit

They tried to send a recording and it bounced. They need it under a specific number today and do not care about codecs — they want to know what to set and whether the recording will still be usable.

Aim at about 18 MB, not 25. Gmail's famous 25 MB ceiling applies to the whole encoded message, and mail attachments travel as base64 text, which is four characters for every three bytes of file. That is a 33% tax before headers, so a 25 MB limit passes a file of roughly 18 MB. For a spoken recording that is a lot of room: at 64 kbps mono, 18 MB holds about 37 minutes. The route to it is three moves in order — cut the dead air, switch a voice recording to mono, then re-encode at a bitrate that suits speech. Most recordings clear the limit after the first two.

The limits that actually apply

Two limits matter and only one is yours. Your provider decides what it will send; the recipient's provider decides what it will accept. The smaller number wins, and you cannot see theirs.

Gmail is the helpful case: go past 25 MB and it puts the file in Drive and sends a link rather than refusing. That is fine for a colleague, and awkward for anyone outside your organisation, who then needs permission on a Drive file to hear a voice memo.

Chat platforms have their own caps and they move. Discord has raised its free per-file limit more than once, and WhatsApp treats an audio file sent as a document differently from a voice note. Rather than trust a number copied from a blog post, drag the file in and read what the client tells you. The arithmetic below works the same whatever the target is.

The arithmetic, which is the whole trick

File size is bitrate multiplied by duration. Nothing else is involved for a compressed format. The useful form of it:

• Megabytes ≈ bitrate in kbps × minutes ÷ 133. A 20-minute recording at 128 kbps is about 19 MB.
• Bitrate to ask for ≈ target MB × 133 ÷ minutes. An hour of audio into 18 MB needs about 40 kbps.

So you can pick the setting before converting anything, rather than encoding three times to find it by trial. The same sum read the other way — how much recording fits in that 18 MB of usable space:

The bottom two rows explain most bounced emails. An interview recorded to WAV is roughly ten times the size of the same interview as a 128 kbps MP3, and a laptop set to record uncompressed will produce a 600 MB file for an hour without ever mentioning it.

Step one: cut the dead air

The cheapest megabytes are the ones with nothing in them. Recordings start before the meeting and stop after it, and a lecture capture often carries five minutes of room tone at each end. Removing them costs nothing at all, because the part you keep is untouched.

The audio cutter here trims without re-encoding wherever the format allows, so a trim is a copy rather than a compression pass. Trimming first also pays off downstream: a 25% shorter file needs 25% less bitrate reduction to hit the same target, which means a better-sounding result at the end.

Step two: mono, if it is a voice

One person talking into one microphone, recorded to two channels, is the same signal stored twice. Downmixing to mono removes an exact duplicate. For an uncompressed WAV that halves the file outright.

For MP3 or AAC the saving is indirect and worth understanding. A stereo file at 128 kbps and a mono file at 128 kbps are the same size — the encoder simply spends every bit on one channel instead of splitting them. The win is that mono at 64 kbps sounds about as good as stereo at 128, so going mono lets you halve the bitrate without hearing the difference. Choose mono and lower the number.

Mono also quietly fixes a common defect: a recording where the microphone only reached one channel, so half your listeners hear silence. If that is your file, downmixing is the repair as well as the saving — audio that only plays in one ear covers what causes it.

Do not do this to music. Stereo imaging is most of what makes a mix sound like a mix, and collapsing it is a bad trade for anything anyone will listen to for pleasure.

Step three: pick a bitrate that matches the content

Speech carries far less information than music, which is why it survives compression that would ruin a song. Sensible floors, mono for the speech rows:

• 32–48 kbps — phone quality. Intelligible, obviously compressed, fine for a note to yourself or for a transcription tool.
• 64 kbps — the usual sweet spot for an interview or a meeting. Clear, natural, and small enough that a 35-minute conversation clears a Gmail attachment with room over.
• 96 kbps — speech with room over, or a podcast someone will hear on headphones.
• 128 kbps stereo — the floor for music, and the point at which most people stop noticing anything is missing.
• 192 kbps stereo — music you care about. Above this the file grows faster than the quality does; 320 versus 192 kbps goes through what blind tests actually show.

Pick the number from the arithmetic, not from a habit. If your 45-minute recording has to fit 18 MB, that is 53 kbps, so 48 kbps mono clears it and 64 does not. The compression tool shows the projected size as you move the setting, which is quicker than converting to find out.

What re-compressing an existing MP3 buys you, and what it costs

Taking a 320 kbps MP3 down to 96 kbps works and gives you the size you asked for. What happens inside: the file is decoded back to samples and compressed again by a second encoder that has no idea what the first one already threw away. The damage accumulates. For a single generation it is nothing like as bad as the folklore suggests, and it is real.

Two things follow. If you still have the original recording, compress from that rather than from an MP3 someone already made — one generation of loss instead of two. And never ask for a higher bitrate than the source had: 128 kbps re-encoded to 320 is three times the size with none of the detail back, because no encoder can rebuild what the first one discarded.

Changing container alone does not help either. An M4A carried across to another container the same codec allows is a stream copy — the audio comes out identical, every compressed byte and every decoded sample, and only the container's metadata is rewritten. It finishes in about a second whatever the length, and the file is the same size it was. Useful for compatibility, useless for an attachment limit.

Doing the conversion without sending the recording anywhere

The converter on this site is FFmpeg compiled to WebAssembly. The browser fetches the engine once, about 32 MB, and caches it for a year; after that the page reads your file off your own disk and works on it in the tab. The recording never leaves your device, which matters more than usual for the things people email — a client call, a recording someone gave you in confidence.

Pick MP3 if you have no idea what the recipient uses, since nothing made in the last twenty-five years refuses it. M4A at the same bitrate sounds slightly better and is safe on Apple hardware or a modern Windows install. The channel setting and the bitrate sit on the same panel, so both saving moves happen in one pass.

When to stop compressing and send a link

Compression has a floor. Below about 32 kbps speech starts to sound underwater, and no setting turns a two-hour board meeting into a 15 MB attachment worth listening to. Reach for a link when any of these is true:

• The recording is over an hour and the recipient needs it in good condition.
• It is music, or anything where the mix matters.
• You are sending several files at once — the limit is on the whole message, not per attachment.
• The recipient is behind a corporate filter that strips large attachments regardless.

Drive, OneDrive, Dropbox and WeTransfer all do this, and Gmail and Outlook offer it at the point where the attachment will not fit. One thing worth doing even then: compress the file to a sensible bitrate before putting it in shared storage. A 600 MB WAV in a Drive folder is a 600 MB download for whoever opens it. A 30 MB M4A of the same meeting is not.

When the recording and the email are both on the phone

If the recording is a voice memo and the message is going out from the same phone, moving the file to a computer to shrink it and back again is a lot of steps. Our Audio Converter app is free on both stores, works offline once installed, does not fetch a 32 MB engine over cellular, and keeps a history so a finished file can be shared again later.

Stated plainly, because it changes the advice: the Android build compresses on the phone. The iOS build's format conversion currently fails and returns an error — its editing tools all work, including the trim and the stereo-to-mono downmix, which are the two steps that matter most here, but on an iPhone use mobile Safari for the actual re-encode.