MP3, AAC (M4A), WAV, FLAC and OGG: which audio format to use for what

You have a recording — a voice memo, an album ripped years ago, the soundtrack of a lecture video — and something is refusing it: a car stereo whose manual lists only MP3, an iPhone that will not open .ogg, a podcast host with an upload cap, an editor that insists on WAV. The sound is the same in every case; what differs is how it is packed, how much was thrown away to make it small, and what each device can decode. This guide gives you the numbers: what each format costs per minute, the bitrates at which lossy formats stop being distinguishable from the original, which format each platform wants, and which conversions lose nothing versus which lose twice.

Three ways to store the same sound

Uncompressed audio (WAV, AIFF) is a list of sample values, so its size is arithmetic: sample rate × bit depth × channels. CD audio is 44,100 samples per second × 16 bits × 2 channels = 1,411 kbps, or 10.6 MB per minute whether that minute holds an orchestra or silence. Lossless compression (FLAC, ALAC) predicts each sample from the previous ones and stores only the small error; the decoder rebuilds the original bit for bit, and the file is 40–70 % of the WAV. Lossy compression (MP3, AAC, Vorbis, Opus) runs a psychoacoustic model that finds what a louder sound would mask and discards it before coding. Files come out 5–15 times smaller than WAV, and the loss is permanent: whatever the model removed is gone from every copy made afterwards.

Mono halves the lossy and uncompressed figures. “Transparent” below means the bitrate above which listeners in blind tests stop telling the encoded file from the original. For MP3 with the LAME encoder that is roughly the V2 preset, 170–210 kbps, with V0 (about 245 kbps) as insurance; AAC-LC gets there around 128–160 kbps and Opus around 96–128 kbps. Above those numbers you pay storage for nothing audible; below them the first casualties are cymbals, applause and reverb tails.

The five formats, one at a time

MP3 — plays on everything, tops out at 320 kbps

MP3 is MPEG-1 Audio Layer III, standardised in 1993 and patent-free since 2017. It is limited to 48 kHz, to mono or stereo, and to 320 kbps. The encoder that matters is LAME: its -V scale runs from 0 (≈ 245 kbps) to 9; V2 averages about 190 kbps and low-passes at roughly 18.6 kHz, while 128 kbps CBR is low-passed at about 17 kHz — which is why 128 kbps sounds dull on cymbals and why that ceiling stays visible in a spectrogram years later. The encoder also adds 576 samples of delay, so a looped MP3 has a gap at the join unless the player honours the LAME tag. Tags and artwork travel in an ID3v2 block ahead of the audio. The MP3 converter uses exactly this LAME V2 preset.

AAC / M4A — the streaming and Apple default

AAC (Advanced Audio Coding) is MP3’s designed successor from MPEG-2 and MPEG-4: sample rates to 96 kHz, up to 48 channels, and better than MP3 at every bitrate — most visibly below 128 kbps, where the HE-AAC profiles let broadcasters run 48–64 kbps stereo. Files almost always hold AAC-LC (low complexity). The container is the important part: .m4a is an MP4 file beginning ....ftypM4A , and it can hold AAC or Apple Lossless (ALAC) — Apple Music’s lossless tier is ALAC in .m4a — so the extension alone does not say whether a file is lossy. A bare .aac is usually a raw ADTS stream with no tags or seek index. iPhones, Android, Windows 10 and later, every browser and YouTube play AAC-LC in M4A. The M4A converter writes AAC-LC at 192 kbps.

OGG — Vorbis and Opus, the royalty-free pair

Ogg is a container, not a codec; every page of one begins with OggS. Inside is one of two Xiph.org codecs. Vorbis (2000) has a quality scale from −1 to 10 where -q 5 gives about 160 kbps and -q 6 about 192; it is sample-accurate, so loops are gapless by design, which together with the absence of licensing is why game engines and HTML5 games settled on it. Opus (RFC 6716, 2012) runs at 48 kHz internally, scales from 6 to 510 kbps, is low-latency enough for calls, and at 64–96 kbps competes with MP3 at twice the rate; it is mandatory in WebRTC and is what WhatsApp voice notes, Discord and YouTube’s WebM streams use. Android, Chrome, Firefox and Edge play both. Apple is the holdout: Music and Voice Memos do not open Ogg, and Safari’s support arrived late and varies by version. The OGG converter produces Vorbis at quality 5, not Opus.

FLAC — the archive format

FLAC (2001) is lossless, open and supported everywhere outside Apple’s own apps: Android, Windows 10, the Files app on iOS, VLC, foobar2000, Bandcamp, most network players and a growing share of car head units. Compression levels run 0–8; the default 5 typically lands pop and rock at 55–65 % of the WAV and sparse classical or speech nearer 40 %. It takes up to 8 channels, 32-bit samples and any sample rate, carries tags as Vorbis comments plus a PICTURE block, and stores an MD5 of the decoded audio in STREAMINFO, so flac -t album.flac verifies a ten-year-old archive bit for bit. The FLAC converter uses that default level 5.

WAV and AIFF — for editing, not for keeping

WAV is Microsoft’s RIFF container around raw PCM — a fmt chunk describing rate, depth and channels, then a datachunk of samples; AIFF is Apple’s big-endian twin. Every editor, DAW, sampler and telephony system wants them because nothing has to be decoded and seeking is arithmetic. Two weaknesses are structural. RIFF sizes are 32-bit, so a WAV cannot exceed 4 GiB — 6.7 hours at CD quality, two hours at 24-bit / 96 kHz — which is why field recorders split long takes. And tagging is a LIST INFO chunk with four-letter keys (INAM, IART), no artwork standard and patchy player support. The WAV converterwrites 16-bit PCM at the source’s sample rate and channel count.

Sample rate and bit depth: the numbers that do not need to be big

A sample rate captures every frequency up to half of itself: 44.1 kHz covers 22.05 kHz and 48 kHz covers 24 kHz, while adult hearing stops between 15 and 20 kHz. 44.1 kHz is the CD and music rate; 48 kHz is the rate of every video and broadcast standard. Bit depth sets the gap between the loudest sample and the noise floor — about 96 dB for 16 bits, 144 dB in theory for 24 — and since a quiet room sits near 30 dB SPL and a loud peak near 100 dB, 16 bits already cover everything you can use at playback. 24-bit earns its size while recording and editing: you can leave 18–20 dB of headroom above your peaks with the noise floor still below the microphone’s own, and every gain change afterwards rounds to a bit nobody hears. DAWs mix in 32-bit float for the same reason.

The rule that bites is not “higher is better” but “do not mix them”. A 44.1 kHz file in a 48 kHz session that is not resampled on import plays 8.8 % fast and about a semitone and a half sharp (48 ÷ 44.1 = 1.088). Pick 48 kHz for anything that meets video, 44.1 kHz for pure music, and convert once, at the start. The converters here keep the source rate and channels, with two exceptions the codec forces: MP3 cannot exceed 48 kHz, so a 96 kHz master is resampled to 48 kHz, and MP3 is mono or stereo only, so a 5.1 soundtrack is folded down to stereo.

What a conversion can and cannot do

Lossy to lossy loses again. An MP3 re-encoded to AAC is decoded — artefacts included — and quantised a second time by a different model; the two sets of decisions do not line up, so the damage compounds. When you must, go up in bitrate, never down. Lossy to lossless restores nothing. FLAC and WAV preserve what they are given; an MP3 converted to FLAC is four or five times larger and sounds identical. Lossless to lossless is free. WAV, FLAC and ALAC convert among each other with no change to the audio, subject only to bit depth — this site’s WAV output is 16-bit, so keep the original of a 24-bit master. The policy that follows: keep one lossless master of anything you recorded or ripped, and derive every MP3, M4A or OGG from it, never from another lossy file.

Loudness is a separate axis that formats do not touch. Podcast hosts commonly quote −16 LUFS integrated for stereo (−19 for mono) with true peaks under −1 dBTP; music services normalise playback to somewhere around −14 to −16 LUFS. None of the converters here normalise — they encode the level they are given — so adjust in your editor first, then convert.

Which format for which job

• Car stereos, old phones, anything with a USB stick: MP3, the one format they all decode. LAME V2 is the right setting; 320 kbps CBR is 70 % larger for no audible change.
• Podcast episodes: MP3 for the feed — every app plays it and every host accepts it — at 64–96 kbps mono for speech or 128 kbps stereo with music, normalised to −16 LUFS first.
• Music archive: FLAC from the CD, WAV or 24-bit master. Tag it, verify it with flac -t, and treat every MP3 as a disposable copy.
• Voice memos: keep the M4A your phone recorded; make an MP3 only for a device that cannot play AAC.
• Game and web assets: Ogg Vorbis for music and loops (gapless, royalty-free); WAV for short effects that must fire with no decode. On the web, ship M4A alongside OGG if Safari matters.
• Video soundtracks: give the editor WAV at 48 kHz and let the export write AAC into the MP4 — what YouTube, iOS and every TV expect. The containers and codecs guide covers what each video container can hold.
• Sending to a person: M4A or MP3, which chat apps and mail clients play inline; a WAV or Vorbis OGG usually arrives as a generic attachment. Gmail caps attachments at 25 MB — 26 minutes of 128 kbps MP3 or 2.4 minutes of WAV; above that, send a link instead.

How to tell what you actually have

The extension is a hint: .m4a may be AAC or ALAC, .ogg may be Vorbis or Opus, and a .flac may have been born an MP3. The first bytes and one ffprobe call settle container and codec.

# first twelve bytes as hex, ASCII alongside
head -c 12 mystery.audio | xxd
#   4944 33..                       "ID3"          -> MP3 with an ID3v2 tag in front
#   fffb / fff3 / fff2 ....         frame sync     -> MP3 with no tag
#   5249 4646 .... 5741 5645        "RIFF....WAVE" -> WAV
#   464f 524d .... 4149 4646        "FORM....AIFF" -> AIFF
#   664c 6143                       "fLaC"         -> FLAC
#   4f67 6753                       "OggS"         -> Ogg: Vorbis or Opus, see below
#   0000 00xx 6674 7970 4d34 4120   "....ftypM4A " -> M4A: AAC or ALAC, see below

# codec, rate, channels, bit depth and bitrate of the audio stream
ffprobe -v error -select_streams a:0 \
  -show_entries stream=codec_name,sample_rate,channels,bits_per_raw_sample,bit_rate \
  -of default=noprint_wrappers=1 mystery.audio
#   codec_name=aac | alac | vorbis | opus | mp3 | flac | pcm_s16le | pcm_s24le

Two traps. file reports the bitrate of the first MP3 frame it sees, which is meaningless for a VBR file — ffprobereads the Xing header instead. And whether a lossless file was ever lossy is invisible to both: look at a spectrogram (Audacity’s track menu, or sox in.flac -n spectrogram -o spec.png). Real recordings fade gradually towards 20 kHz; a hard, flat ceiling at 16–17 kHz is the low-pass of a 128 kbps MP3 encoder, and one at about 19.5 kHz is LAME V0. Either way the file is a transcode, whatever the label says.

How the converters on this site work

All five converters share one engine: FFmpeg compiled to WebAssembly, loaded into the page (a 10 MB download the first time), with a fixed preset per tool and no options. Drop one file or several and conversion starts at once; the input accepts audio/* and video/*, so an MP4, MOV, WebM or MKV works as a source and its soundtrack is what gets converted. The audio is decoded and re-encoded in the tab’s memory and never sent anywhere.

Every run is ffmpeg -i input -vn <preset> output. The -vn drops the picture from a video file and, because FFmpeg treats embedded artwork as a one-frame video stream, drops cover art too; text tags are carried across where the target has somewhere to put them. A single file downloads as name_output.mp3 (or .m4a, .ogg, .flac, .wav) and appears in a player on the page. Several files convert in turn and download together as converted-mp3.zip, each keeping its base name; a file that fails is skipped and named in the summary (Could not convert: lecture-03.mkv.) while the rest continue. The whole input is read into memory as one buffer — fine for a two-hour lecture, but the ceiling is roughly 2 GB per file. A single failure shows Conversion failed. The file may be unsupported, corrupted, or contain no audio track. — and the last clause is the usual cause, a screen recording made with no microphone. For the untouched soundtrack of a video rather than a re-encode, see the extracting audio guide.

Common mistakes

• Converting MP3 to FLAC or WAV to “upgrade” it. Five to ten times larger, same 17–19 kHz ceiling. The only reason is a tool that refuses compressed input — and then it wants WAV.
• 320 kbps for speech. A lecture has no cymbals to protect. 64 kbps mono is more than a voice needs: a 90-minute lecture is 43 MB instead of 216 MB.
• WAV as an attachment.Ten minutes of CD-quality WAV is 106 MB, over four times Gmail’s limit, and a phone shows it as a file rather than playing it. Send M4A or MP3; hand WAV only to someone who will edit it.
• OGG to someone with an iPhone. It will not open in Music, Voice Memos or the Messages preview. Vorbis is for engines, browsers and Android; people get M4A.
• Converting an ALAC M4A “to M4A”. The output is AAC at 192 kbps: same extension, no longer lossless. Check codec_name with ffprobe before touching anything you believe is an archive, and take lossless sources to FLAC.
• Mixing 44.1 and 48 kHz in one project. Beds at 44.1 kHz in a 48 kHz timeline are resampled on the fly or, in engines that do not resample, play 8.8 % fast. Convert every asset to the project rate once, before import.