After recording my weekly piece of audio data this evening very early this morning, I tried to get spectrum plots using Audacity in order to compare what the non compressed / losslessly compressed audio and the compressed audio that would later on go on air sounded like. Here are the results for:

  • Non compressed audio (PCM / WAVE);
  • Losslessly compressed audio (FLAC);
  • Lossy compression:
    • MPEG-1 Layer III (MP3) at 256 kbps;
    • MPEG-1 Layer II (MP2) at 256 kbps, the codec we use to store our audio material;
    • Ogg Vorbis at quality 6 for comparison purposes.

The original material is the WAVE file. All files are stereo 48000 Hz audio. Spectrum analysis done only on the first 21.8 seconds of audio.

Nice plots using gnuplot:

100 Hz ‒ 24000 Hz spectrum

Closer look at the high frequencies (13 kHz ‒ 24 kHz — not a log scale):

13 kHz ‒ 24 kHz spectrum

As always, MPEG has a frequency cutoff under 20 kHz (this may be audible for young people), but remember this is 256 kbps MPEG... at 128 kbps, cutoff is around 15 kHz... Ogg Vorbis has a higher cutoff (around 21 kHz). WAVE / FLAC have no cutoff, but the audio data is quiet background music + my voice, and I don't think I can produce a sound higher than 22 kHz ;) ).

For the rest of the audio spectrum, nothing special to my (tired) eye.