Exported Mp3s Don't sound as good in Linux as does Windows

Audacity 2.3.0

I’m hoping to revive this discussion.

What makes this all very difficult for me is I have over 8,000 mp3s and the majority of them was made to the 128 CBR as mentioned. Otherwise I would just turn down the treble and be happy.

I made a thread about this 2 years ago. But I got busy and was unable to follow through with all the necessary tests. What I ended up doing is just giving up on it. But now it’s become more necessary that I figure out what is causing this.

My old thread is here: https://forum.audacityteam.org/t/constant-bit-rate-export-problem/43636/6

Way back when I started playing around with mp3 files, I noticed that as I lowered the constant bit rate this would happen. 1) The Highs would become a little slurred but the bass for the most part would remain the same or sound lower as the extreme highs would get removed. (That’s the best way I know how to describe it)

I’ve always chosen my CBR by determining how much quality per how much disk space / how easy is it for the mp3 player to load up the file.

I’ve always liked 128 CBR with Sampling rate : 44.1 KHz.

What I always did is buy the highest quality mp3s from Amazon and then use Audacity to export them to 128 CBR. I’m fully aware that this reduces quality. But on the equipment I run these mp3s on, I never noticed enough of a difference to bother my ears.

Today I worked with the following songs from Chatterbox on Windows 7

01 - Torque.mp3
02 - Empty.mp3
03 - Fallen.mp3
04 - Spine.mp3
05 - Internal.mp3
06 - Soul Scum.mp3
07 - Divide.mp3
08 - External.mp3
09 - Epignosis.mp3
10 - Sunshine.mp3

Yesterday I worked with the same songs on Linux Mint 18.3, using the same version of Audacity. The music hurts my ears. When I down grade these files, the sound becomes higher. (The Bass, The Mids, and The Highs don’t stay like the original) The highs become a little higher, or some of the bass is lost. I’m not entirely sure which is happening… The problem with my hearing is that I’m more sensitive to highs then I am to lows.

I’m using a Sandisk Sansa Clip + running the Rockbox firmware. I reset all the settings back to default. So that the EQs were for sure set to flat. I’m using the Koss UR40 head phones, which is my favorite head phone model that I’ve used since the early 2000’s I also ran this same test in my car, and my car stereo gives me the same effect on my hearing. In fact in my car I needed to lower the treble by a lot to get it to sound like something I would call “normal”. What is really odd is I have trouble telling the difference in sound quality on the computers using the same head phones I use with my mp3 player. But I don’t think that is saying anything. Often times computers have access to better codecs. Where as your mp3 players, you are pretty much stuck with whatever happens to be in the firmware.

In my last thread I was asked about the lame versions.
Linux Mint 18.3

Version 3.99.5+repack1-9build1 (xenial)

Ubuntu Developers <ubuntu-devel-discuss@lists.ubuntu.com>

Depends: libc6
Depends: libmp3lame0
Depends: libsndfile1
Depends: libtinfo5

Windows 7

I'm running Lame 3.99.3

I do not have FFmpeg installed

Linux Mint I do have FFmpeg installed

Version 7:3.4.4-1~16.04.york0
  • Things are a little more complex then what I wrote here. I’ve experimented with WinFF on Linux Mint 18.3 which gave me 160 CBR out of the box. And they were really high. I’ve also been experimenting with 192 CBR in Audacity. But it always seems like what I get on Linux Mint 18.3 is a little higher sound frequency then what I get on Windows 7.

I’m not able to reproduce the problem.
Try converting this clip on both Windows and Linux, load the two versions into your media player, then listen to them side by side.

We may not be able to reproduce the problem. Because a lot of this does depend on our hearing and audio equipment. But lets have some fun with this and see what we can do.

I think I can hear just a very very very slight difference between the two. Linux Mint being just a tad higher then the Win7. The best way I can describe it is it’s almost like when I was a kid back in the late 80’s I could walk into a computer room and hear the really old CRT monitors high frequency that is almost outside of normal hearing range.

I think if this was all just in my head, then after a long period of time, and I forget what computer made the file on, I shouldn’t notice a difference. And yet I’ve been at this since 2016 and I can always tell. There is just something that always seems a little off.

I also made some samples of what I’ve been working on. First thing I did is export a 8 second section to WAV and then I exported it to 128 on the Linux Mint computer and then I exported the same thing again on Win7. The music is copyrighted but I can’t see any reason why I would get into trouble just for 8 seconds. I would prefer to share exactly what I’m working on since I think you can notice the “problem” in more intensive forms of music.

Feel free to post a few seconds of what you are working on.
The two Mahler MP3s sound identical to me.

Fortunately we can eliminate much of the subjectivity by looking at measurements.
For example, as expected, these are not identical, but they are extremely similar:

Decoders don’t vary as much as encoders. Compression (encoding) is the tricky part where the encoder has to figure-out what data to throw-away. Virtually all encoders are running LAME in the background so they can be the same too, but depending on the interface it may be tricky to get the exact same settings on two different systems.

And, some decoders are integer based which can cause clipping on the peaks. MP3s often go over 0dB so they can clip (distort). I’ve never heard of a case where the clipping was audible (the quality probably won’t improve if you lower the volume before MP3 compression) but that might be what you’re hearing and why you’re hearing a difference on different playback systems. If the (decoded) MP3 goes over 0dB and you play it at “full digital volume” you’ll clip the DAC but you can avoid that by reducing the volume digitally before it hits the DAC.

I’ve always chosen my CBR by determining how much quality per how much disk space / how easy is it for the mp3 player to load up the file.

I’ve always liked 128 CBR with Sampling rate : 44.1 KHz.

CBR uses the same number of bits to record silence or simple sounds as complex sounds. VBR and ABR are “smarter” and they try to allocate those bits moment-by-moment to the parts of the track where they are needed most so you generally get better quality for a given file size. ABR (average bitrate) allows you to set the actual average bitrate. With VBR (variable bitrate) you choose a quality setting and the overall bitrate varies so it’s harder to predict file size.

Also, if you’re not using Joint Stereo some of the information is being encoded twice. Again Joint Stereo is “smarter” and it makes better use of the bits. (The Joint Stereo process itself is lossless… It just helps to make the lossy part of the compression more efficient/effective.)

What I always did is buy the highest quality mp3s from Amazon and then use Audacity to export them to 128 CBR. I’m fully aware that this reduces quality. But on the equipment I run these mp3s on, I never noticed enough of a difference to bother my ears.

There is “damage” every time you compress. A 128kbps MP3 made from a higher-bitrate MP3 may not sound as good as a 128kbps MP3 ripped directly from a CD or made from a lossless original. Damage accumulates even if you re-compress to the same or higher bitrate.

At least I now know I’m not crazy. :unamused:

I tested the Mahler Mp3s in my car. At first I thought I could tell a slight difference and then after a few listens I couldn’t tell anymore. They are so close to each other that even if I’m hearing something my brain made up for the difference. Kinda like when you’ve been playing around the equalizers a little too long, and can no longer tell if your improving or making it worse.

Through Linux Amazon won’t allow me download the CBR of my choice. They require I use the Amazon app to do that. I wonder if they are tapping into files they already have stored at a lower quality or are they just doing the same thing I’m doing by downgrading them. Do you know?

I’ve done this for so long I wasn’t expecting to have this problem. As I said before I’m aware that I’m loosing quality. One theory of mine was to downgrade them to 192 instead of 128. The other day I wanted to test out 44 mp3s so I used WinFF on linux mint 18.3 to downgrade them to 160. What came out, I finally decided I couldn’t stand to listen too. That was what gave me the extra nudge I needed to start this thread up.

Just about all the music I like to listen too is over dynamically compressed and clips a lot. Apparently that is what they did in production. Therefor I never tried to change it. Again, if I can’t hear the clipping distortion mixed in with the artistic side of what the artist is doing then I have no reason to care.

I’m happy to make you guys clips. These are the latest two I’ve been testing with. I can’t tell a difference between these in my car, currently. However since I already have them made up for this thread I’ll go ahead and post them. As far as I know all my music is still on Amazon and I’m allowed to re download them. I think what I will try is install the amazon app on my windows 7 computer and try downloading at 128 just to see if there is any difference between that and what I’m making on my own.

Update on Amazon Music APP for PC.

I just tested the Amazon Music App for PC. I installed it onto Windows 7. The current version does not seem to support the ability to give you a choice in music quality. The last time I used this App would of been, maybe 3 years ago or so. I know that back then it did give me the option. Here is a link the current Amazon Music App: https://www.amazon.com/gp/help/customer/display.html?nodeId=201377740