Hello All,

I have been attempting to decode some MP3s to WAV and/or burn the MP3s onto an Audio CD. Fairly straight forward stuff.

One thing that I have noticed is that these MP3s appear to contain audio in the frequencies over 16khz (as expected) that is lost when either decoding to WAV or burning to CD.

I have tried using the fraunhofer PRO codec; winamps own wav writer; Nero; Foobar 2000; Razorlame and many more. All of these programs appear to fail to decode the frequencies over 16khz.

I have even tried transcoding in Nero and Foobar but still the same result.

Does anybody have any experience of this? Can anybody explain what is happening and give me a method of resolving the matter?

Regards,

Marc

If the MP3 is at 128kbits/s or 160Kbits/s , it most probably does have the cutoff frequency of 16Khz so it wouldn't be there to begin with.

If it were an MP3Pro file, you would have a cutoff at 10Khz (if not decoded properly) and not at 16Khz so i doubt it is the case.

Reading again your post, you just seem to notice that you lack these frequencies, but haven't found a way to validate this.
If you use foobar, you can see it with the spectrum analizer vis. plugin (what you are seeing is what is being reproduced). If you use Nero, you can aswell use the Wave editor and use the spectrum view.

To coclude, you either are imaginating it ( what you'll read in these boards called "placebo effect" ), or the hardware ( amp, speakers... ) where you play your CD are better (better response).
There are no known problems with these software when decoding MP3's.

Ps. If you *really* think you're having a problem, maybe giving the description of how you do it and uploading a sample of the .mp3 in question could help.

Foobar spectrum analyser and winamps analyser shows that the 128kbps MP3 source does have frequencies above 16khz but then when I use the same software to play the decoded wav file the spectrum is flat at 16khz and above.

Just for the record - I am not sure of the exact cutoff frequency - it is roughly 16khz.

Its very strange and I have to say that I have barely noticed any actual difference in the sound of the music itself but I know enough to know that the high frequencies, although indistinct to the ear on their own, have an effect on ones perception of the audio.

Could the encoder be the source of this problem. I have a feeling that lame wasnt used in this case..... (terrible IMO).

Marc

I don't know about foobar, but the Winamp spectrum analyser is very misleading. I certainly wouldn't use it to judge the spectral content of a file.


You'll almost certainly find that (as long as you're using a decent encoder - and, to be honest, even if you're using a very bad one!) that the decoded content is basicaly what is in the file. The encoder didn't include anything above 16kHz. Spectral analysis of most mp3 files reveals that the encoder cuts info above 16kHz much more aggresively than below 16kHz, even in the absence of a hard 16kHz low pass. This is very common in mp3 files, even lame ones.

Too late to do anything about the files now. Don't worry about it if you can't hear it. But use a better encoder (or format!) in the future if you can.

EDIT: Though the --alt presets maintain content above 16kHz when it matters, the mp3 format itself means this is quite inefficient.

Cheers,
David.

Thanks for the advice and comments.

I only ever use lame when working with MP3 for the reasons you suggest. I am sure that the person who encoded this stuff didnt.

I use MP4 myself but thats another story.

I'll try a test with sound forge spectrum analyser later - that should show what the source contains without doubt. Anybody know of any other software that will allow a more detailed study of the source?

Is there any way of identifying the encoder used on MP3 files?

Encspot will tell you what was used (and i think the frequecy range as well). Generally it doesn't matter. I've tried clipping the high stuff at 16k and never noticed the difference (and I'm younger then most around here).

If you by "frequency range" mean the lowpass frequency used in the encoding process, Encspot will only be able to show that if Lame was used and a Lame tag was written to the file during encoding.

All th posts above seem to have solved your problem, I am just wondring why you choose to blame the decoding process for the loss of high frequencies. When your favourite player plays an MP3 on your PC, it decodes it to PCM format (same as WAV files) and sends the PCM data to the sound card. Soundcards (with a few rare exceptions) don't have hardware on them that understands MP3 (or any other format besides PCM).

What this means is that any comment of the form "Decoding MP3 breaks the sound" are meaningless. Besides a few broken decoders, they should all sound very similar if not identical. Some people have managed to ABX decoder problems in the past - but this always seems to be based on subtle things (like dithering) rather than serious problems (like a frequency cutoff).

What he meant is that the spectrum looks different when the MP3 is played and when the wav is played.

Precisely!

All of the things mentioned above I know about. My problem lies in the fact that there appears to be activity in the range above 16khz in the original MP3 files. This activity no longer exists when the MP3 is transcoded/decoded to WAV format.

The original DEFINATELY contains sound above the 16khz range so why and how is this information lost when the MP3 is converted to a WAV file?

Thank you for your feedback and input - it is much appreciated - has anybody tried to replicate this problem?

Question: How do you know it contains it? Can you cut the parts you can hear and post them?

I will post a picture showing the frequency range for both the MP3 and the WAV created from the MP3 file - as soon as I get back home.

As for hearing the sounds above 16khz.... I havent tried a highpass filter yet. I doubt that I would hear the sounds over 16khz to be blunt but I know that they have an effect on the perception of the lower frequencies - the ones that really matter.

The analyzer definitely display something, but it doesn't mean that there is something. First, it can be noise. Analyzers can display as low as -150 dB. In here, quantization or dither noise appear very strong. It can also be the leaking of a lower frequency analyzed through too short a window, or with a window that leaks (Hanning, square, triangle...). But there shouldn't be a difference between the files in the second case. There might be one in the first case.

Using Winamp disc writer plugin should sort this out : it will write the MP3 playback to wav; dither, volume, eq and DSP included. The resulting wav should have the same treble as your MP3.



The tests that we are aware of indicate otherwise with consumer or audiophile gear : ]Slides from the AES convention in Banff on intermodulation distortion in loudspeakers and its relationship to "high definition" audio. , from http://world.std.com/~griesngr/
96 vs. 48 or 44.1 kHz sampling --> scientific test, perhaps here is the 1. listening test !
More links in the FAQ...
People at George Massenburg forum, including myself, confirmed Griesinger's test.

The only success in showing that inaudible frequencies affect audible ones is the experiment by Ooashi et al. In order to make inaudible frequencies affect audible ones, they used an experimental tweeter with a frequency response up to 100 kHz !

Well, most of the tests claim that HF frequency cutoffs are below the 16kHz range... which is totally bull; I ABX-ed in a hurry the 17kHz&18kHz lowpass here: http://www.kikeg.arrakis.es/lowpass/
(M-Audio Revo+Senn HD600); only 5/8 for the 19kHz one (5/5 then 5/8) - should be impossible for me as I can hear tones only up to 19.5kHz at normal listening volume.

A file: C:\002 Chenoa_17KHz.wav
B file: C:\001 Chenoa_17KHz_lowpass.wav

Start position 00:01.2, end position 00:03.3
10:30:30 1/1 p=50.0%
10:30:39 2/2 p=25.0%
10:30:44 3/3 p=12.5%
10:30:50 4/4 p=6.2%
10:30:56 5/5 p=3.1%
10:31:01 6/6 p=1.6%
10:31:07 7/7 p=0.8%
10:31:16 8/8 p=0.4%
10:31:21 test finished

edit: ABX'ed the 18kHz lowpass:

WinABX v0.42 test report
11/12/2003 10:37:12

A file: C:\Chenoa_18KHz.wav
B file: C:\Chenoa_18KHz_lowpass.wav

Start position 00:01.0, end position 00:03.1
10:37:45 1/1 p=50.0%
10:37:55 2/2 p=25.0%
10:38:02 3/3 p=12.5%
10:38:11 4/4 p=6.2%
10:38:17 5/5 p=3.1%
10:38:25 6/6 p=1.6%
10:38:30 7/7 p=0.8%
10:38:37 8/8 p=0.4%
10:38:38 test finished

No, try a lowpass filter better You'll hear the 16kHz+ only content for sure given the right volume, but hearing it in the actual music is another thing. I can hear up to 19.5kHz (test tones) but only able to distinguish 18kHz lowpasses... 24 years old, btw.