Since I have been doing a bit of spectral analysis of music to learn about the frequency distribution of the content, I started looking into the changes made by digital music compression. I took short (20-30 seconds) sections of two songs and made charts. I ran the full songs through various compression techniques to get the file sizes (this was previously stated incorrectly). The results are below (file sizes are given as a percentage of the original for reference).

Cold, Cold Heart - Norah Jones ‘Come Away with Me’

Original WAV file
MP3 128kbps CBR 9% of original
MP3 256kbps CBR 18% of original
MP3 VBR preset-extreme 16% of original
WMA 128kbps CBR 9% of original
WMA 135-215kbps VBR 19% of original
WMA Lossless 44% of original

Sweet Child O’ Mine - Guns N’ Roses ‘Appetite for Destruction’

Original WAV file
MP3 128kbps CBR
MP3 256kbps CBR
MP3 VBR preset-extreme
WMA 128kbps CBR
WMA 135-215kbps VBR
WMA Lossless

As you can see, there is an obvious loss of high frequency information in all forms of compression except for the WMA lossless. Some, like MP3 256CBR, only lose a small amount, but others, like WMA 128CBR, lose a significant amount (everything above 16kHz in this case).

Of course, these images don’t show enough detail to see other potential impacts of compression. Perhaps we’ll visit that another time.

Spectral charts done using Adobe Audition 1.5. MP3 compression done using LAME 3.92 via CDex. WMA compression done using Windows Media Player 10.

---