A creator hears the published clip change.

The obvious suspects are the microphone and the level.

They are not the only suspects.

A spoken clip, a music bed, and a live conversation ask an export to do three different jobs.

One needs to play anywhere.

One needs enough data to keep its detail from thinning out.

One needs the delay short enough that two people can talk without waiting for each other.

A recording can be clean when it leaves the room and still be wrong for the job it gets at export.

Bitrate is how much data a second of audio receives. It sets the room a lossy codec has to work in.

It doesn’t turn the recording up.

MP3 and AAC lean on masking: when a louder sound sits beside a quieter one, the encoder can drop what you were unlikely to hear anyway.

Which is why lossy compression doesn’t damage every recording the same way.

Speech gives a codec plenty of places to economize. A dense music bed gives it fewer. A sharp transient gives it almost none.

Around 128 kbps, MP3 and AAC are audibly compressed on transient-heavy material.

Around 192 kbps, you are at the common streaming default.

Around 320 kbps, most listeners stop hearing the difference on most content.

Those aren’t three rungs on a ladder toward a master.

They are three allowances for a delivery file.

A number that looks generous in an export menu is still describing a portable copy — not a recording that survives being encoded again.

A 320 kbps MP3 makes a sensible compatibility copy, and MP3 remains the safe answer when you don’t know what will play it.

It is still not where the work should live.

AAC strikes a different bargain, reaching the same perceptual quality as a top-rate MP3 with noticeably less data.

That makes it a good fit when consumer delivery matters more than preserving every sample.

Neither choice repairs a file already compressed for the wrong job.

Upload a lossy MP3 somewhere the platform re-encodes to AAC and you have started a second lossy process.

The first encoder removed information. The second can’t bring it back.

It listens to the altered signal and finds something else to remove.

So keep a source apart before you make any delivery file.

WAV keeps the audio uncompressed. FLAC keeps the same sound in appreciably less space.

Either one preserves the samples a later remaster or conversion will need.

They are not oversized versions of the delivery copy.

They are the version the next delivery copy comes from.

Choose the sample rate for the work as well. 44.1 kHz is the consumer-music standard; 48 kHz is the standard for audio that travels with video.

Those numbers describe how often the waveform gets measured.

They say nothing about bitrate, and they rescue nothing.

Bit depth answers another question again. Recording at 24-bit rather than 16-bit gives you headroom while tracks are still being mixed.

Headroom helps before delivery.

It doesn’t turn a thin music file into a master afterwards.

Then there is allocation.

CBR hands every second the same amount of data, which means an orchestral swell and a silence are treated as equals.

VBR moves data toward the difficult passages and away from the easy ones, and can reach the same perceived quality in a smaller file.

ABR varies the same way but aims at a single average across the file.

Fixed allocation suits a stream whose bandwidth has to stay predictable. Variable allocation suits a finished recording that should let the material decide.

Don’t ask either of them to solve a live conversation.

AAC gathers audio into frames before it encodes them, and that gathering costs time before the network has added any of its own.

Fine for a published song.

Slow for two people trying to interrupt each other.

Opus was built for that other job.

It works in far shorter frames, holds intelligible speech at very low bitrates, and still handles complex music well.

That combination — low delay, speech, and music in one codec — is why real-time voice runs on it.

Opus isn’t a newer MP3.

It is an answer to a question MP3 was never asked.

So start with the version that has to survive.

Keep it in WAV or FLAC.

Make MP3 or AAC copies when compatibility and ordinary delivery are the point.

Reach for Opus when the point is a conversation happening now.

And don’t let the portable copy replace the source just because the portable copy sounds acceptable today.

The microphone captures the signal.

The export decides what the signal is allowed to keep.

A clear recording can still be given the wrong job at export.