My uncle was a DJ before and he's really into MP3 details. He says that a song should sound good and also look good when viewing it on the computer. Also 19000hz is not nessecary so we must change that to 16000hz. When I say the MP3 must look good, here's a screenshot of the wrong look:

http://img520.imageshack.us/img520/1808/untitled1bc8.jpg

If you look at those purple lines, he believes that it is background noise and in order to have perfect MP3 we have to remove it. By doing that we have to turn the sound down from this:

http://img165.imageshack.us/img165/7811/untitled2xq7.jpg

to this:

http://img501.imageshack.us/img501/2479/untitled3ta9.jpg

Ive now applied a -9dB Cut, set it to 16000hz and saved as MP3 with Frauenhofer since LAME doesn't let us do that much in Adobe Audition. Also he wants me to save it in CBR since thats compatible with everything and VBR is not.

Now the spectral frequency display shows (the first image I posted) this:

http://img179.imageshack.us/my.php?image=untitled4dq3.jpg



Ok now I have explained his way but now comes my questions:

Is he right about the 160000hz? He says a human ear can't hear more then that or something slightly above so going from 19000hz to 16000hz shouldnt make a difference. My first concern about this was those calm songs where the singers sing a very high tune (remember Im a total newb and know almost nothing), I was worried that their high tones might be cut off but he told me no. So is he right?

Now for the Amplify settings. Do I really have to turn down the sound? He says that all the songs we get from the web are twisted when we try and play them with really high volume so he says we should turn it down to -9db cause this fixes that problem and now when we play it with very high volume it isnt ruined anymore. Is he right? Will it sound the same when I play it OR do we actually lose something when we turn it down?


Wow writting all of this at 5AM w00t. Thanks guys and I kneel before you wise MP3 knowledge. Also I apologize for my english, when it comes to writing technical stuff Im afraid it looks like Im still in kinder garden but I imagine you all get what I mean ^^.

Firstly, loading an MP3 in an editor then saving (effectively reencoding) it is one of the worst things you can do. If you start with a WAV file, ie. directly ripped from the CD, then it's fine. As for lowpassing, and MP3 encoders, it's widely accepted at most bitrates LAME does the best job (for music), and that the LAME presets provide the optimum settings for lowpass etc. See here for details: http://wiki.hydrogenaudio.org/index.php?ti...cal_information

Secondly, what do you mean by 'songs from the web' ?
There is a huge variation in the quality of anything you download from the internet, if it's from a reputable source it should be of a consistent quality, but not necessarily high quality. Unfortunately, at the moment, illegal sources (ie. torrents) provide by far the best quality in downloaded music, although a few legal sites are starting to offer lossless downloads now, along with higher bitrate MP3 and AAC.

what I mean by 'songs from the web' is songs that are ripped uploaded on the web.

From your answer I really didnt understand much. So what your saying is that taking a d/l MP3 into Adobe Audition and then saving it is wrong?

High voices are nowhere near 16000 Hz - more like 4000 Hz (babies crying) for the fundamental. Many sounds have overtones at much higher frequencies, and many noise-like sounds and sharp clicks and percussions have a broad frequency spectrum.

Most people struggle in blind tests on music to tell whether a 16000 Hz (16 kHz) low-pass filter has been applied, though younger people could be more sensitive to high frequencies, albeit that these tend to be hidden in the presence of lower frequencies (the difference between real music and test signals).

A good MP3 encoder will automatically adjust the lowpass to achieve best quality. See the Wiki (link above) and LAME for recommended settings. VBR is aimed at constant quality, with -V5 --vbr-new considered to be very good (sometimes distinguishable from the original, but usually not annoying differences ~ roughly 130 kbps on average, with just over 16.75 kHz lowpass) and -V2 --vbr-new (roughly 190-200kbps) considered to be 'transparent' (completely indistinguishable on most music for most listeners).

The vertical bars all the way up to 22050 kHz on the spectrogram are caused by clipping. This may or may not be audible, but can easily be fixed on MP3 files without having to decode and re-encode the MP3 (this is called transcoding and degrades quality, and is what happens if you open an MP3 in Audition and then save it to any lossless format, such as MP3).

The easiest solution is probably mp3gain. This free software can adjust the "global gain" for each frame of the mp3 without re-encoding it. If you apply a gain of -3 dB you'll fix most mp3 clipping with no loss of quality, just slightly less loudness. It's also possible to apply "Max No Clip Gain". My preferred solution is to use mp3gain for equal perceived loudness across albums, you might choose Album Gain, which allows you to shuffle play your albums without annoying variations in loudness, and which reduces clipping greatly at the same time. (In practice I usually apply this using foobar2000 before encoding to MP3, though it can do the same thing that mp3gain does in modifying the MP3 directly)

in the most polite way possible:

-you don't listen with your eyes, it doesn't matter what it looks like.

-frequencies higher than 16k are definatley audible, that's why they are there in the first place. whether or not it's nessecary depends on the quality your after. but taking away = losing quality.

-yes a lot of music these days is either mastered with too much compression and clips, or people have used normalization and gain controls to clip/compress the resulting mp3's. But -9db is overkill, the more you "turn it down" the less depth there is, ie. difference between loud and soft. and dependant on how good your audio playback equiptment is, such a large cut will only reveal more of the noise floor and hiss/interference in your signal line.

-transcoding (mp3-wav-mp3) has already been addressed but it will NEVER improve quality, it's impossible. avoid at all costs.

-cbr is more compatible than vbr, but in the same way low bitrate mp3's are more compatible than higher bitrate mp3's. compatibility is a trade-off. these days vbr is widely accepted. '-V0' (vbr) or '-b 320' (cbr) are both very close to the best quality mp3 you can get using the current lame encoders.

While you're technically correct, I disagree with warning against this sort of reduction.

In practice 16-bit audio is way more than adequate even for music mastered to "proper" average loudness to preserve dynamics, let alone music mastered to be as loud as possible. 14-bit was considered ample by Philips when designing the CD with Sony, and 13-bit was deemed ample by the BBC for digital transmission of highly dynamic music on Radio 3 to FM transmitters. Losing 2 bits is -12 dB, losing 3 bits is -18 dB increase in noise floor. Given that today's music is recorded at 12 to 18 dB louder than most audiophile recordings, there's probably room to drop these very loud recordings by 24 to 36 dB (fading the bottom 4 to 6 bits into the noise) before the noise floor becomes a problem when played back at normal to loud volume.

-9dB is negligible in this context, and applying Replay Gain (which tends to be of this order of magnitude for modern music) is also benign in real-world conditions, even without taking good care to dither the audio.

I wouldn't want to scare away someone new to the field who is thinking of reducing volume in their files, particularly when mp3gain and other Replay Gain methods bring such a worthwhile benefit with no commonly audible downside.

Ok thanks guys this really helped me.

One more thing I was wondering about. Ive got this program called MP3 Frame which removes all the tags in a MP3 file but by doing so do I lose bits or such of the song? It can remove the following tags:

ID3v2 is a tagging format that appears at the front of the stream meaning the whole stream has to be rewritten if a tag edit changes its length.

ID3v1 is widely supported on digital audio players (DAPs) and MP3 CD players (e.g. in-car) but has limited field lengths.

A RIFF header is small and should only be present on a WAV file, but you can embed MP3 inside a WAV wrapper (not common these days)

Lyrics3 is a tag containing song lyrics

Unknown header and footer data may include other tagging standards like APEv2 tags (footer). The option "Unknown header data" might remove the LAME/Xing 'tag' which is written as a valid but silent MP3 frame to allow it to be ignored by MP3 players (but if it's a valid frame, it shouldn't). Those that understand it can obtain information from it (including gapless playback info) and will seek better in VBR files. You can test whether the LAME tag is removed by writing one (encode a track using LAME -V2 --vbr-new for example, check the Properties in foobar2000 to see enc_padding and enc_delay gapless information, then remove unknown header data and check again, reloading info from file). Foobar2000 can also rebuild the MP3 stream if there is a problem.

If you do start removing tags and have any fear of removing something you might want, make a backup copy first so you can restore the original files.

Anyway, editing/removing tags should not affect the music data, unless there's as problem with the tag editor. The exception is if there's any extra metadata that you use for playback (like Replaygain tags or gapless info).

Ok thanks. You guys have really helped.