Convert AMR to MP3

Convert AMR voice recordings from old Nokia and Android phones to MP3 in your browser. Free and private, with no upload and no size limit.

🌐 Español

Drop your file here (.amr)

🔒 Private by design: your files are processed locally in your browser and never uploaded to any server.

Written and maintained by Last updated

AMR is a telephone codec, not a music codec

Adaptive Multi-Rate was standardised by 3GPP in 1999 for one specific job: carrying a human voice across a GSM mobile network using as little radio bandwidth as possible. It became the speech codec of the 2G and 3G era, which means it was already inside every handset, which in turn is why phone manufacturers used it for voice memos, voicemail and MMS audio without writing anything new.

The “adaptive” part refers to its eight bitrate modes, from around 4.75 up to 12.2 kilobits per second, which the network could switch between mid-call as reception changed. A recorder app just picks one and stays there. The wideband variant, AMR-WB, standardised separately as G.722.2, doubles the sample rate and runs between roughly 6.6 and 23.85 kilobits per second; it is what “HD Voice” marketing referred to.

None of this is compression in the sense that MP3 is compression. AMR uses code-excited linear prediction, which does not store the waveform at all. It fits a mathematical model of a human vocal tract to the sound and transmits the parameters of that model, and the decoder reconstructs speech from them. That is why it is so astonishingly small and why it falls apart on anything that is not a person talking.

8 kHz sampling and the ceiling it puts on the recording

Narrowband AMR samples the signal 8,000 times a second. Sampling theory sets the highest reproducible frequency at half of that, so nothing above roughly 4 kHz exists in the file, and in practice the usable band is narrower still. Human speech intelligibility mostly lives below that line, which is exactly why the telephone system settled on it a century ago and never moved.

The consequence for you is straightforward. Whatever is in an AMR recording, it is mono, it is band-limited, and it was encoded by a speech model. Converting it to MP3 changes the packaging so that ordinary software will open it. It cannot widen the frequency range, restore the missing high end, or separate two people talking over each other.

That is worth saying because plenty of converter sites imply otherwise with words like “enhance” and “high quality output”. The honest framing is that this makes the recording usable, not better.

The bitrate ladder here tops out at 192 kbps on purpose

An MP3 bitrate ladder normally tops out at 320 kbps, the highest rate the format defines. This one does not, and the reason is written into the module: pushing a phone-quality voice recording up to a music-grade bitrate would produce a much larger file without adding any real fidelity.

So the Audio quality rungs are scaled to the source. 128 kbps (recommended) is the default and is already generous for band-limited mono speech. 192 kbps (highest quality) exists as headroom if you want the encoder to have no excuse at all, typically because the recording is going somewhere formal. 64 kbps (smaller file) is the practical choice when you have a folder of call recordings to archive and every one of them is a person talking into a handset.

If you want the general reasoning rather than this page’s specific ladder, choosing a bitrate for voice versus music sets out what speech actually needs. The short version is that speech needs far less than people assume, and AMR proves the point at a twentieth of these rates.

The MP3 comes out several times larger than the AMR

Expect the file size to go up. A ten-minute AMR memo can be under a megabyte; the same recording as a 128 kbps MP3 is closer to ten. Nothing has gone wrong and no setting will make MP3 match a codec built specifically for the telephone band.

What you get in return is a file that opens. AMR support outside phone recorder apps is close to nonexistent: desktop players skip it, video editors do not list it, cloud storage previews show nothing, and transcription and documentation systems reject the extension on sight. A voicemail from a relative, a recorded meeting or a saved call is stuck wherever it was made until it becomes an MP3.

Modern messaging apps solved the same problem with a better codec, which is what OPUS to MP3 deals with, and it has the same size behaviour for the same reason. If text rather than audio is what you actually need from an old recording, Transcribe Audio runs speech recognition on your own machine, though its accepted list has no .amr in it, so convert here first and hand it the MP3. The editing pages on the audio tools hub work the same way round: they take the MP3, never the AMR.

Converting a phone voice memo

  1. Drop the .amr file onto the box above, or click Choose a file and select it.
  2. Leave Audio quality on 128 kbps (recommended) for a normal voice recording, or drop to 64 kbps (smaller file) if you are working through a batch.
  3. Click Convert AMR to MP3. A progress bar and a Cancel button take over while FFmpeg decodes and re-encodes.
  4. Download the MP3, which keeps the original filename with .mp3 in place of .amr.

Frequently asked questions

My phone saved the recording as .3ga rather than .amr. Can I use it here?

Not on this page, because the accepted list is .amr only and anything else is refused with an unsupported-type message before conversion starts. A .3ga is AMR audio inside a 3GP container rather than a bare AMR stream, so it needs a demuxer for that container. Renaming the extension is not a reliable fix.

Why does 64 kbps sound the same as 192 kbps on my recording?

Because there is very little in the source for the higher settings to describe. An AMR narrowband file carries mono speech sampled at 8 kHz and encoded at somewhere between roughly 5 and 12 kilobits per second, so even the lowest rung offered here has several times the capacity needed to reproduce it faithfully.

Someone sang on my AMR recording and it sounds strange. Can that be fixed?

No, and it was not the conversion. AMR encodes by modelling a human vocal tract rather than by storing the waveform, which is what makes it so efficient for speech and so poor at everything else. Music put through it comes out warbling and metallic, and that damage is baked into the file before you get here.

Will the volume of an old phone recording improve?

Not by itself. The conversion changes the container and the codec and leaves the level exactly where it was, so a quiet recording stays quiet. Raising it is a separate step, and the MP3 volume booster here is the tool for it, at the cost of a second encode once this conversion has produced the MP3 it needs.

Can I run several call recordings through in one go?

Not in a single pass. This page handles one file at a time, and after a conversion finishes the dropzone ignores new files until you click Process another to clear the result. The upside is that the encoder stays loaded in the tab, so the second and later files start much faster than the first.

Related tools