PCM, WAV and AMR: Three Formats on Three Different Levels

Start here: don't let the acronyms scare you

When I started out, seeing PCM, WAV and AMR in a row made my head spin. They read as textbook jargon, somewhere off in engineering land.

Then I spent a few years actually cutting podcasts, pulling voice memos off phones and dealing with call audio, and it turned out none of them are exotic — they aren't even the same kind of thing. Most people line them up as three competing audio formats and immediately get confused.

So let's go outside-in. PCM is the raw data at the bottom. WAV is a container someone put it in. AMR is a completely different road: a codec built specifically for narrowband scenarios like phone calls.

Once that hierarchy is clear, picking a format stops being guesswork.

1. PCM: the ancestor of all audio quality, and the dumbest of them

PCM is Pulse Code Modulation. Sounds like a communications exam question; the idea is almost silly in its simplicity.

Picture sound as one continuous wavy line — an analog signal. What PCM does is take a knife and slice that line at fixed intervals, measure the height of each slice, and write the number down.

Those three moves are the whole thing:

1. Sampling: how often do you cut? That's the sample rate
2. Quantization: how precise is each measurement? That's bit depth
3. Encoding: store those measurements as binary

Do all three and analog sound becomes a pile of numbers. That's PCM.

Two parameters, and you can hold your own in conversation

Sample rate, in Hz — how many cuts per second.

• 22050 Hz: voice grade. Speech is intelligible
• 44100 Hz: the CD standard. Music lives here
• 48000 Hz: the default for video and film

Nyquist's theorem, which has an intimidating name and a simple claim: to reproduce a frequency of f, you need to sample at 2f or more. CD picked 44.1 kHz because human hearing tops out near 20 kHz, doubled, with a little margin. So no, that number wasn't invented.

Bit depth — how many bits per sample.

• 8-bit: coarse, that old telephone and chiptune texture
• 16-bit: CD grade, plenty of dynamic range for normal people
• 24-bit: studio mixing territory, lots of headroom, hard to clip in post

The canonical example: CD quality = 44100 Hz / 16-bit / two channels. Keep that as your yardstick and every other format becomes easy to place.

PCM's personality: pure and dumb

Pure because nothing is thrown away. It is digital audio in its raw form — as lossless as lossless gets, and it defines what the ceiling even means.

Dumb because it doesn't compress. Do the math: 44100 × 16 bits × 2 channels works out to about 1411 kilobits per second, which is 30 MB or more for one song. Record a two-hour podcast in PCM and you've burned through gigabytes.

Worse: PCM strictly speaking isn't a file format at all, it's naked data. Save it as .pcm and the player has no idea whether to read it at 44.1k or 48k, so you get noise or something that sounds like the tape is on fast-forward. That's exactly why the next one exists — WAV staples the instructions to it.

2. WAV: PCM with a shell, and everybody's safe choice

WAV (Waveform Audio File Format) came from Microsoft and IBM, built on RIFF (Resource Interchange File Format). The structure is two chunks:

[ header ]  <- the instruction sheet
[ data    ]  <- the actual PCM, or some other codec's output

The header is full of metadata: sample rate, bit depth, channel count, how long the data block is, what codec is in use. A player reads the header, sees 44.1k / 16-bit / stereo, feeds it at that rate, and it sounds right.

Important: WAV does not automatically mean lossless

This misconception deserves a whiteboard.

WAV is a container, not a codec. Most of the time it holds PCM, which is lossless, but it can also hold compressed material like ADPCM. So calling every .wav lossless isn't strictly correct — it depends on what's inside.

In practice, 99 percent of WAV files are PCM and everyone treats them as lossless. Fine. Just know why.

Why did it win?

One word: compatibility.

Audition, Reaper, Premiere, FFmpeg, Windows Media Player, your handheld recorder — nobody refuses WAV. Editors use it as the interchange format: record to WAV, edit, export MP3 or FLAC. That pipeline hasn't really changed in decades.

• Upside: lossless, top-tier quality; compatibility that borders on absurd; a complete header that post tools read cleanly
• Downside: fat. It carries PCM, so it's as fat as PCM. WAV is for working, not for messaging — mix those up and you're the person with a 400 MB file that won't send

3. AMR: a narrow road built for phone calls

The first two preserve quality. AMR inverts the goal — it exists for speech and for saving bandwidth.

AMR is Adaptive Multi-Rate, and the adaptive part is the whole idea. It watches your voice and moves the bit rate around: compressing hard during quiet or simple passages, opening up when detail appears. Save whatever it can without making you unintelligible.

3GPP specified it for mobile networks (GSM and WCDMA), so the codec in your phone calls is usually AMR or a relative. The .amr extension is what you saw on old Android recordings, early WeChat voice messages and MMS audio.

How much does it save?

AMR normally hops between 4.75 kbps and 12.2 kbps. Compare that with PCM's 1411 kbps — a factor of over a hundred.

A minute of speech might be a few dozen kilobytes in AMR and over ten megabytes in WAV. That gap is why it could carry a conversation across 2G-era bandwidth.

The cost is real, and it's large

AMR is lossy on top of lossy. The algorithm is tuned for the human voice, so it throws away the harmonics and high frequencies that music depends on. Store a song in AMR and you get "music heard over a phone call" — muffled, dull, instruments collapsed into one another.

So its lane is obvious:

• Calls, voice memos, very low bandwidth → AMR is great
• Music, post-production, listening → stay away

One practical annoyance: desktop players and editors support AMR inconsistently. FFmpeg handles it; some tools refuse to import it. If someone hands you an AMR file to edit, convert it first:

ffmpeg -i input.amr output.wav

4. All three side by side

	PCM	WAV	AMR
What it is	Raw digital audio (a coding scheme)	A container for audio	Speech-specific compression
Lossy?	No	Usually not (depends on contents)	Yes
Size	Huge	Huge	Tiny
Quality	The ceiling	The ceiling (when it holds PCM)	Intelligible speech, nothing more
Support	Poor — you must supply parameters	Excellent	So-so, mostly mobile
Typical use	Studio masters, low-level development	Editing interchange, sample libraries	Calls, phone recordings, MMS

See the shape of it? PCM and WAV are family — one naked, one dressed. AMR came over from a different sport. Comparing WAV and AMR on sound quality is like comparing a thermos and a water bottle on which one holds boiling water better. Different jobs.

5. What I actually pick, by job

Recording a podcast, voiceover, or anything you'll edit

Set the recorder to 44.1 kHz / 16-bit (or 24-bit) WAV. Don't record to MP3 and don't touch AMR — that's planting a problem, and three edit passes later the audio is porridge.

Storing the intermediate

Keep the master as WAV or FLAC (FLAC is lossless compression and saves some room, as covered in the last piece). Never delete the original; that's your option to re-cut later.

Sending it out

AAC for platforms, 320 kbps MP3 for your own distribution. Insisting on WAV at this stage isn't rigor, it's self-sabotage.

Quick phone memos and call-related apps

AMR is legitimate here: small, fast, cheap on data, and fine for speech. If you're building an app, read up on AMR-NB versus AMR-WB — narrowband and wideband are not interchangeable.

Writing code or doing audio development

You'll meet PCM directly — AudioTrack, Core Audio, FFmpeg's AV_SAMPLE_FMT_S16. Sample rate, bit depth and planar versus interleaved need to be exact or you get noise. And the WAV header? Writing the 44 bytes yourself to wrap PCM into a playable file is a well-known interview exercise and worth doing once.

Finally

Underneath it all, these three represent three different trades:

• PCM says: I want maximum fidelity; file size is not my problem
• WAV says: I want fidelity that every program accepts; fine, I'll be fat
• AMR says: I only need speech to arrive; cut everything else

Understand what each one chose to give up, and you'll never pick wrong again.

And next time someone tells you WAV always beats MP3, you can quietly add, "depends what codec is inside the WAV." People who know will notice 😎

When it's time to convert, Audio Converter moves recordings between the common formats with the bit rate under your control, and Video Converter pulls a track out of video. Everything runs in your browser: no uploads, no retention. Your privacy is fully protected.

Now I'm going to turn that AMR voice memo into WAV and get on with the edit. May none of your files be too big to send 🫡

Common questions

Is WAV always better sounding than MP3?

Not automatically. WAV is a container, not a codec. Usually it holds PCM, which is lossless, but it can also carry compressed formats like ADPCM. So the precise answer is "it depends what's inside the WAV," even though in practice nearly every WAV you meet holds PCM and treating it as lossless is fine.

Why does audio saved as .pcm play back as noise?

Because PCM is bare data with no header declaring sample rate, bit depth or channel count. The player doesn't know whether to read it at 44.1 kHz or 48 kHz, so you get static or something that sounds sped up. Add a WAV header of 44 bytes spelling out the parameters and any player can read it.

Which is smaller, AMR or MP3?

AMR, by a wide margin. It adapts between roughly 4.75 and 12.2 kbps, while CD-grade PCM runs about 1411 kilobits per second — over a hundred times more. A minute of speech can be a few dozen kilobytes in AMR versus ten-plus megabytes in WAV. The trade is that it's tuned for voices, so music comes out muffled and flat.

Someone sent me an AMR file to edit. Now what?

Convert it to WAV first. Desktop editing tools support AMR inconsistently and some refuse to import it, so FFmpeg is the reliable route: ffmpeg -i input.amr output.wav. Once it's WAV, Audition, Reaper and Premiere all handle it normally.

Note: some of the material here was produced with AI assistance.

---

---