Spectrum of Musical Genres

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Administrator
Joined 2004
Paid Member
I thought you might find this interesting. It's the spectral content of music visualized. You can see the tonal balance of different types of music.

The graphs below are FFT analysis of different music genres. The files analyzed for the this test are the average of approximately 112 different recordings in each genre. I averaged those recordings and plotted the spectrum.
Note that the graphs do NOT show how loud the tracks were mastered, because all tracks were normalized before averaging them. That allows for the overlay on one of the plots. The plots just show tonal balance. You can see the relative levels of Bass, Midrange and Treble for each style of music.

  • Rock section is almost all Classic Rock. Zepplin, Foghat, AC-DC, Janis Joplin, Eagles, Billy Idol, etc.
  • Jazz tracks are mostly mid-century Bebop or Straight ahead. Dizzy, Miles, Coltrane, Mingus, Art Blakey, Horace Silver, Dave Brubeck, etc.
  • Classical section is all orchestral, not chamber or choral music.
  • Opera is a wide selection of mostly 19th century works.
  • Hip-Hop is whatever I had. Eminem, Dr Dre, and others.

Let me know what you think.
 

Attachments

  • spectrums-1.png
    spectrums-1.png
    34.2 KB · Views: 1,771
  • spectrums_offset.png
    spectrums_offset.png
    45.6 KB · Views: 1,891
  • Like
Reactions: 1 user
Administrator
Joined 2004
Paid Member
Well, I wondered about that. The Hip-Hop looks very uniform, like pink noise. Rock is similar thru the midrange. My guess is that it's the wall of sound producing, with limited dynamics and no empty spaces that does it. Just a guess, tho. More graphs coming.
 
Administrator
Joined 2004
Paid Member
We should test for flat response, we don't want a curve like this! But this should give you an idea of how much energy there is in different parts of the spectrum.
What I see is that rock & jazz are similar, looking like brownian noise that is high passed at 60 Hz 4th order. Hip-Hop is about the same, but high passed at 50 Hz.
And there is the roll off at the top, which in classical and opera is more striking.
 
Pano,

I did a similar study three years ago and posted the results on TechTalk.
Here is a summary graph. It is pretty similar to yours except I didn't normalize so that I could see something about the dynamic headroom allowed compared to the average. I also used 1/6 octave averaging, so I don't get the -3 dB/oct rolloff that you get with a straight FFT. Pink noise comes out flat. The spectral averaging also hides the 1/12 octave peaks from the chromatic scale that you see.

attachment.php


The details are here:

Musical genre specific average spectral content - Techtalk Speaker Building, Audio, Video Discussion Forum

Marc
 

Attachments

  • music spectra.png
    music spectra.png
    128.1 KB · Views: 2,901
Last edited:
Thanks for sharing. 100 recordings averaged for five genres is lots of work.

Does this tell us (Very very very broadly) about Audio hardware requirement ? Like 2way speakers driven by moderate power amplifier will satisfy 90% need of audio specifications required ? Can average dynamic range of all music genres be calculated from graphs ?
Thanks again.
Regards.
 
Administrator
Joined 2004
Paid Member
Well, maybe. The main idea behind my tests was to find the shape of the music curve. For rock and Jazz it looks pretty similar. Bass rolls off under 60 Hz at 24dB/octave. The top end rolls off above about 80Hz at 4.5 dB/octave.

Pink noise with its 3dB/octave roll off is typically used for loudspeaker testing, for technical reasons. Brownian noise, which rolls off at 6dB/octave is often said to be more like the music spectrum than pink noise. I've always found it to be in between the two, as my average plots show.

Over in the Beyond the Ariel thread there was some testing of compression in bass drivers using pink noise. But that got me to thinking that sine sweeps and pink noise always sound far too bass heavy during tests. They don't sound like the tonal balance of music. Of course sine wave sweeps also contain far more top end energy than pink noise or music, I've popped a few tweeters that way. :(

Maybe it would be better to do some of our testing with a signal that more closely mimics the musical spectrum than current test signals. It should not be our only test signal, but it could be handy. I've been able to filter white noise and sine sweeps to very closely follow the Rock/Jazz spectrum - and hope to try using them in tests. I can post those signals here, if there is interest. It's funny to listen to the filtered sine sweep, it's far from linear, but does not sound as odd as I thought it would.
 
Member
Joined 2014
Paid Member
Very interesting. Now member Cask05 considers the deviation from pink noise to be a mastering artifact and re-eqs his recordings. There is certainly a good argument for the LF rolloff as most small speakers just wouldn't handle the bass. For HF I'm not so sure, but I haven't scanned the literature for the frequency balance of orchestral music at either the recording or listening positions.

Anyway Pano, :up::up: for doing this.
 
Administrator
Joined 2004
Paid Member
Yes, the spiky graph for classical and opera is an unexpected result. I don't know if it has to do with tuning, dynamics, or something else.

And to answer Hearinspace's question, I don't know why there is such a difference between my classical plot and Marc's. Is it the recordings chosen, the method of averaging? I just don't know. Maybe I need to try a different 100 tracks to see how repeatable the results are.
 
I am guessing that the difference in the classical recordings is about the particular pieces selected. Also, these are accumulated averages. If I were to do this again, I would take both average and peak measurements, because peak is what sets how much amplifier is needed. I know that some of the classical had powerful content below 40 Hz; it just was intermittent. Other genres have more continuous bass lines or percussion and push the average up.

An interesting variation would be finding the peak of the average power with time constants varying with frequency: long for heavy woofer voice coils, and much shorter for tweeter voice coils. This could give useful information along with the absolute peak that would indicate peak amplifier needed vs. the RMS power handling of different drivers.
 
Does this tell us (Very very very broadly) about Audio hardware requirement ? Like 2way speakers driven by moderate power amplifier will satisfy 90% need of audio specifications required ?
This tells us what frequency range is needed for music reproduction.

Can average dynamic range of all music genres be calculated from graphs ?
Thanks again.
Regards.
Not from these graphs.
 
Administrator
Joined 2004
Paid Member
Not from these graphs.
Correct. Since my tracks were all nornalized and mixed together, dynamic information is lost.

However, as I've posted over the years, a typical early CD mastering level was for the RMS value of the music to be 18 dB below peak. That makes for a fairly dynamic recording. Most classical is even more dynamic with 22 to 26dB of headroom. Another typical level is 16dB below peak which is used on a lot of recent jazz and vocal recordings. I've seen Metallica recordings with an average of 10dB below peak, and I suppose that new recordings are even worse.

But that's another topic, and has been covered in abundance in other threads.
 
Administrator
Joined 2004
Paid Member
To answer the question about peak values vs average values, here is another look. These are the same rock tracks, but this time the yellow indicates peak values. These peaks are not weighted as to how often the occur, just that they did occur.
Any way, the peaks seem a little flatter thru the bass than the average values, but they two are not so far apart.
 

Attachments

  • rock peaks.png
    rock peaks.png
    50.5 KB · Views: 425
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.