Detecting mp3-sourced audio is not entirely simple or 100% reliable, but there are a couple of reasonably easy things to try.

The basic approach to detecting mp3-sourced audio is to run spectral and frequency analysis on some of the wav files. That sounds worse than it actually is! Here's why:

Many audio fans already have Exact Audio Copy (EAC) installed, and can use that for frequency analysis (see a link to specifics below). In addition, there are many other audio editors that have these features.

It's easy enough to rip one or two of the wavs from a CDR, or decode some files from mp3 to wav, to do the analysis. By the way, if you are not going to use these wavs to burn, you don't need to worry about doing a perfect, secure rip.

You can also decode an mp3 file to wav, to do this type of analysis and see what different files or encodes look like. This is a good way to check the quality of any mp3s you might have, and should encourage to replace (or at least not circulate) mp3s encoded with obsolete encoders or at lower bitrates. Many programs will convert mp3s to wav files (including most of the all-in-one programs like MusicMatch). mkwACT is a compact, useful utility which can decode mp3s. With all of these, though, you may need to be careful to ensure it is using a good quality decoder.

Frequency Analysis

wav. In a good wav, if you do a "frequency analysis" which plots the db vertically and the frequency horizontally, you will see the line trail off as it goes from 10-22 khz. Here is an example, using Prove It All Night from Winterland Night:

Figure 1: Frequency analysis of an original wav file

mp3. If you decode an mp3 to wav and do a frequency analysis on the wav, you will often see a sudden drop to much lower volume (db) level at around 17-19khz (or on old, bad mp3s, you will see it at much lower frequencies). MP3 most significantly affects the higher frequencies. (High frequencies take the most bits to encode and are what we are least able to hear, so to varying degrees you will see the compression affect the high frequencies.) Note how this drops off at around 18-19 khz:

Figure 2. Frequency analysis of an mp3-sourced file

Note that you can see far more differences than you hear, because MP3 is a "perceptual" encoding system based on what humans actually hear. In fact, in all likelihood you would not be able to hear the differences between the wav in Figure 1 and the mp3-sourced (LAME VBR) wav used for Figure 2.

Unfortunately, detecting mp3-sourced boots isn't always this clear-cut. For example, sometimes the effect is less dramatic in the high frequencies, like this one from a 256k bit-rate constant bit rate (CBR) mp3:

Figure 3: Frequency analysis of another mp3-sourced file

The spectral analysis in Part 2: Spectral Analysis will provide more clues to help the analysis of more difficult source analyses.

(Also, looking at Figure 2 and Figure 3, you might think the track in Figure 3 will sound better because the effect of mp3 compression appears less severe than in Figure 2. However, in fact they are roughly comparable in sound quality - both very good, and the VBR uses fewer bits, which is the point of mp3. These two mp3s would likely each be better or worse than the other in different ways, with the VBR better in almost all of the ways that matter. For example, this kind of graph does not really tell you how well the mp3 is capturing the important mid-range frequencies. So, this only illustrates again that looking at audio is in no way a substitute for listening to it - the purpose of this exercise is to detect source issues, not evaluate quality.)

In addition, you can't always be sure what accounts for an unusual frequency profile. It may be that the original source has issues. In particular, mini-disc (MD) recordings will show an even more pronounced effect, as they also use a compression approach roughly similar to mp3 but with more compression. The following figure shows an MD recording transferred to wav (but never encoded to mp3) and illustrates the even more severe compression which MD recordings undergo, which affects the frequencies from around 13-14khz and up:

Figure 4: Frequency analysis of a mini-disc sourced file

You may see similar results if the original source was a radio or television broadcast, or a streaming web file.

There are some other illustrations of frequency analysis at this link. Look first at the "Non-MP3 Sourced Show" example, then compare the others.

---

To find out more, continue on to Part 2: Spectral Analysis

To see how to use EAC for frequency and spectral analysis, click here.

Send any constructive suggestions to contact at audiohub.org

Revised December 14, 2002