fileformer
search⌘KtranslateENexpand_moreupload_fileConvert
Converters8 file families · 44 direct conversions
What changes in a conversionarrow_forward
home/tools/audio-tools

Audio only goes down the ladder. Pick your rung carefully.

Every time sound is squeezed into a smaller format, some detail is thrown away for good. Converting back up makes a bigger file, never a better one. Tell us what you have and what it's for; we'll show which rung to stop at and the tool that gets you there.

formatdetail kept1 song
WAVuncompressed · 16-bit
40 MBoriginal
FLAClossless · zipped
24 MBlossless
M4A · 256kAAC, as sold by Apple
7.7 MBtransparent
MP3 · 320kthe highest MP3 setting
9.6 MBtransparent
OGG · 160kVorbis, used by Spotify
4.8 MBvery good
MP3 · 128kold download standard
3.8 MBgood
OPUS · 32kvoice notes and calls
1.0 MBspeech only
Sizes for a 4-minute stereo song. "Detail kept" is how much of the original a trained listener could tell apart, not a precise measure.
it's for
archiveArchiveheadphonesEveryday listeningphone_iphoneiPhone & Apple MusiceditEditing & DJingpodcastsPodcast uploadmicVoice memos
stop at this rungMP3 · 320kMP3 at 320k sounds identical to the original for almost everyone and plays on every device ever made.
you’ve picked the recommended rungOutlined on the ladder. Click another rung to compare.
Convert to MP3MP3 · 320k · ≈ 9.6 MB per songarrow_forward
~/inside-a-digital-sample

Sample rate sets the highest note. Bit depth sets the quietest one.

Digital audio measures the sound wave thousands of times a second and rounds each measurement to the nearest available step. Below, your browser does exactly that to a pure tone, then measures the rounding error it produced. Push the tone past half the sample rate to hear it fold back into a different note, or make it quiet at a low bit depth to watch it vanish. Every setting can be played and downloaded as a real WAV file.

TONE1 kHz
SAMPLE RATE
8 kHz11.03 kHz22.05 kHz44.1 kHz48 kHz96 kHz
BIT DEPTH
4-bit8-bit16-bit24-bit
TONE LEVEL
0 dBFS, full scale-20 dBFS-40 dBFS-60 dBFS
Add dither (±1 step of random noise before rounding)
play_arrowPlay 1.5 sdownloadWAV · 66.2 KB
Played at −12 dB to protect your ears. WAV has no 4-bit format, so the download stores 4-bit values in 8-bit samples.
- the real wave- what gets stored● samples3.00 ms shown · faint lines = available steps
Nyquist limit22.05 kHzyou hear1 kHzmeasured SNR23.8 dBtheory at full scale25.8 dBnoise floor-24.0 dBFS1 min stereo WAV5.3 MB
graphic_eqThe rounding error is only 23.8 dB below the tone, so you hear it as gritty distortion. Dither would turn it into steadier hiss. The theory figure is 6.02 × bits + 1.76 dB, for a full-scale sine.
~/beyond-converting

Cut it, shrink it, mix it.

Three tools that change the sound itself rather than its format. Drag the length to see what each would do with a file that long.

recording length42 min
~/how-loud-is-it

Streaming services turn every track to the same loudness. Measure yours first.

Loudness isn't the peak level. It's measured in LUFS, using the ITU-R BS.1770 method: the signal is filtered to roughly match how the ear weights frequencies, averaged in 400 ms blocks, and silent or very quiet blocks are gated out. This meter runs that algorithm in your browser on one of three generated clips or on your own file, which is decoded locally. It then shows what each platform would do to the level.

Loud pop mastersynth mix, limited hardPodcast voicesynthetic speech, peaks −6 dBFSOrchestral crescendo−40 to −6 dBFS over 10 sLoud pop master · 10.0 s · stereo · 97 of 97 blocks above the gates
integrated loudness-3.5LUFS
sample peak-0.3 dBFSThe single highest sample.peak to loudness3.3 dBBelow about 8 dB usually means heavy limiting.loudest 3 seconds-3.2 LUFSShort-term loudness at its highest.loudest moment-2.4 LUFSMomentary loudness: one 400 ms block.
momentary loudness over 10.0 s, scale −60 to 0 · dark = gated out (below -13.5 LUFS)
play_arrowPlay as it is, first 8 s
platformtargetgainnew peak
SpotifyTurns quiet tracks up only as far as a −1 dB peak allows, on the default setting.-14−10.5 dB-10.7 dBplay_arrow
YouTubeTurns loud uploads down. Never turns quiet ones up.-14−10.5 dB-10.7 dBplay_arrow
Apple MusicSound Check adjusts in both directions.-16−12.5 dB-12.7 dBplay_arrow
Apple PodcastsThe recommended level for spoken audio. Nothing is adjusted for you.-16needs −12.5 dB-12.7 dBplay_arrow
EBU R128 broadcastEuropean TV and radio deliver programmes at this level.-23−19.5 dB-19.7 dBplay_arrow
ATSC A/85, US TVThe US broadcast practice that the CALM Act made mandatory.-24−20.5 dB-20.7 dBplay_arrow

Play each row to hear the level that platform would use. Turning a track up is capped so the preview never clips. A real platform either limits the peaks or refuses the extra gain. Peaks are sample peaks: the true peak between samples can be up to about 0.5 dB higher.

~/every-audio-converter

17 converters, sorted by what they do to quality.

Going down the ladder saves space and loses a little detail. Across it changes nothing you can hear. Going up never adds detail, but makes files editors and CD burners accept.

southDown the ladderSmaller files, a little detail lost. The most common kind of conversion.
eastAcross, no lossLossless to lossless: bit-for-bit the same sound, just a different wrapper.
northUp, for compatibilityBigger files editors and CD burners accept. No detail comes back.
movieOut of a videoKeep just the soundtrack of a video file as an MP3.
~/reference

Ten audio codecs, tested in the browser you're using.

The last two columns are asked live, through canPlayType and MediaRecorder. “Transparent at” is the rough stereo bitrate where most listeners can't tell the result from the original in blind tests. It depends heavily on the encoder and the music.

codecstandard · yearfilestypetransparent atplays hererecords
MP3MPEG-1 Audio Layer III · ISO/IEC 11172-3 · 1993.mp3lossy≈ 192-256 kbit/sprobablyno
AAC-LCISO/IEC 13818-7 · 1997.m4a .aaclossy≈ 128-192 kbit/sprobablyyes
HE-AACISO/IEC 14496-3 · 2003.m4alossybuilt for 32-64 kbit/sprobablyno
OpusRFC 6716 · 2012.opus .ogg .webmlossy≈ 96-128 kbit/s; 24-32 for speechprobablyyes
VorbisXiph.Org · 2002.ogglossy≈ 160-192 kbit/sprobablyno
FLACRFC 9639 · 2024 (format from 2001).flaclosslessalways: ≈ 50-60% of WAVprobablyno
ALACApple · 2004, open source 2011.m4alosslessalways: ≈ 50-60% of WAVnono
PCMRIFF WAVE · Microsoft/IBM · 1991.wav .aiffuncompressedalways: 1,411 kbit/s at CD qualityprobablyyes
AC-3Dolby Digital · ATSC A/52 · 1995.ac3lossy384-640 kbit/s for 5.1nono
E-AC-3Dolby Digital Plus · A/52B · 2005.ec3lossy≈ 256-768 kbit/s for 5.1 and Atmosnono

Recognise a file by its first bytes

WAV52 49 46 46 .. .. .. .. 57 41 56 45“RIFF”, file size, then “WAVE”
AIFF46 4F 52 4D .. .. .. .. 41 49 46 46“FORM”, size, then “AIFF” (big-endian PCM)
FLAC66 4C 61 43“fLaC”, then a STREAMINFO block
MP349 44 33, or FF FB / FF F3 / FF F2an ID3 tag, or a frame sync with 11 bits set
AAC (ADTS)FF F1 or FF F9a 12-bit sync word
M4A.. .. .. .. 66 74 79 70 4D 34 41 20“ftypM4A ”: an MP4 file with only audio
Ogg4F 67 67 53“OggS”, then OpusHead or \x01vorbis
WebM audio1A 45 DF A3EBML header with DocType “webm”
AMR23 21 41 4D 52 0A“#!AMR” and a newline: old phone voice memos
CAF63 61 66 66“caff”: Apple Core Audio Format

From WAV to Opus

1991WAVMicrosoft and IBM define RIFF WAVE for Windows 3.1 multimedia.
1993MP3MPEG-1 Layer III is standardised, and takes over music sharing within a decade.
1997AACMPEG-2 AAC beats MP3 at the same bitrate; Apple later makes it the iTunes format.
2001FLAC 1.0A free lossless codec that shrinks CD audio to about half its size.
2002Vorbis 1.0Xiph.Org releases a patent-free lossy codec in the Ogg container.
2010EBU R128European broadcasters switch from peak levels to loudness measured in LUFS.
2012OpusRFC 6716 merges Skype’s SILK and Xiph’s CELT into one codec for speech and music.
2024FLAC RFCRFC 9639 makes FLAC an IETF standard.

All audio tools 10