When I am browsing HSK test files (level 6) here (http://www.chinesetest.cn/godownload.do), I found that the decompressed mp3 and wmv is actually larger than the rar file. I'm very surprised since rar is lossless and mp3/wmv is lossy and lossy usually compresses better than lossless. Why might this happen?

Output of unrar -lt:

UNRAR 5.71 freeware      Copyright (c) 1993-2019 Alexander Roshal

Archive: /private/var/root/Desktop/level-6.rar
Details: RAR 4

        Name: level-6.wma
        Type: File
        Size: 26237479
 Packed size: 17275823
       Ratio: 65%
       mtime: 2010-03-18 15:35:19,000000000
  Attributes: ..A....
       CRC32: 950CFA63
     Host OS: Windows
 Compression: RAR 3.0(v29) -m3 -md=4M
UNRAR 5.71 freeware      Copyright (c) 1993-2019 Alexander Roshal

Archive: /private/var/root/Desktop/HSK/H61001.rar
Details: RAR 4

        Name: H61001.mp3
        Type: File
        Size: 86448468
 Packed size: 40408917
       Ratio: 46%
       mtime: 2011-02-18 10:40:02,000000000
  Attributes: ..A....
       CRC32: 7C1319AB
     Host OS: Windows
 Compression: RAR 3.0(v29) -m3 -md=4M

Output of ffprobe:

Input #0, mp3, from '/private/var/root/Desktop/HSK/H61001.mp3':
    album           : δ֪³ªÆ¬¼¯ (2010-9-13 8:32:01)
    title           : ÇúÄ¿ 1
    track           : 1
    genre           : δ֪
    artist          : δ֪ÒÕÊõ¼Ò
    TLEN            : 2162986
    id3v2_priv.AverageLevel: \x0b\x04\x00\x00
    id3v2_priv.PeakValue: \x1a9\x00\x00
  Duration: 00:36:01.11, start: 0.000000, bitrate: 320 kb/s
    Stream #0:0: Audio: mp3, 44100 Hz, stereo, fltp, 320 kb/s
Input #0, asf, from '/private/var/root/Desktop/HSK/level-6.wma':
    WMFSDKNeeded    :
    DeviceConformanceTemplate: L1
    WM/WMADRCPeakReference: 19841
    WM/WMADRCAverageReference: 1995
    WMFSDKVersion   :
    IsVBR           : 0
  Duration: 00:36:10.67, start: 0.000000, bitrate: 96 kb/s
    Stream #0:0(chi): Audio: wmav2 (a[1][0][0] / 0x0161), 44100 Hz, 2 channels, fltp, 96 kb/s

ll (ls -G -lh)

total 630704
-rw-r--r--@ 1 root  staff    82M  2 18  2011 H61001.mp3
-rw-r--r--@ 1 root  staff   700K  2  7 16:15 H61001.pdf
-rw-r--r--@ 1 root  staff    39M  2  7 16:23 H61001.rar
-rw-r--r--@ 1 root  staff   729K  2  7 16:19 HSK-2012.xls
-rw-r--r--@ 1 root  staff   113M  2  7 16:20 HSK6.7z
-rw-r--r--@ 1 root  staff   1.2M  2  7 16:19 HSK6.pdf
-rw-r--r--@ 1 root  staff    16M  2  7 16:23 level-6.rar
-rw-r--r--@ 1 root  staff    25M  3 18  2010 level-6.wma
drwxr-xr-x@ 7 root  staff   224B  3 25  2016 六级录音

Download link:

  • One guess, mp3 is a compression algorithm that is based on fragments of time, with only a small amount of back- and forward-looking in order to keep the decompressor memory and CPU requirements low. Sections of repeated audio (or silence) are likely to result in similar blocks of output being repeated and therefore result in improved subsequent compression by other compressors. You do not decode the entire mp3 to play it as that would result in a need to store up to 40MB for about 3 minutes, fine for high-end devices, but impossible for low end microcontrollers.
    – Mokubai
    Commented Feb 7, 2020 at 10:53

1 Answer 1


It looks like both files are compressed with a constant bit rate (CBR). This will result in a fixed file size no matter what the content of the audio files is. If you're using CBR, 1 minute of silence will compress to the same size as 1 minute of music. Further compression with RAR will pick up on the repetitive pattern of the silence and compress the MP3 file further.

You can also chose to use a variable bit rate (VBR), which will compress silence to a much lower bitrate and thus filesize.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .