Head-Fi.org › Forums › Equipment Forums › Sound Science › I'm bummed :(
New Posts  All Forums:Forum Nav:

I'm bummed :( - Page 2

post #16 of 23
Quote:
Originally Posted by dizzyorange View Post

 

My assumption was that converting lossy to WAV doesn't change the sound.  But in case that's not true, here are the FLAC and mp3 files:

 

http://wikisend.com/download/110086/cityhall.flac

http://wikisend.com/download/218368/cityhall.mp3

 

Converting to WAV will affect the sound, because MP3s are stored using floating point values. If you are encoding to a lossy format, you also need to reduce the gain before encoding in order to avoid intersample clipping. I'm not sure of any programs/encoders that account for this.

You will also find that the better audio players such as JRiver will decode lossy formats to 64-bit internally, rather than 16-bit.
So it is possible to introduce audible artifacts that are not a result of the compression, but of the way the file was encoded.
post #17 of 23

Decoding to 64-bit internally is completely pointless with mp3. First of all, the input is usually 16 bits, secondly, most decoders do not even come close to 32 bits (most do around 16 undistorted bits) and thirdly, floating point output usually has to be quantized to 16 or 24 bit integers clipping anything above 0 dBFS in the process.

 

=> Converting to WAV will not affect the sound.


Edited by xnor - 8/27/13 at 1:00pm
post #18 of 23
Thread Starter 

I downloaded my own sample clips again today and found myself struggling to tell them apart.  After about 10 minutes I could hear the difference again, but only with the advantage of instant A/B switching and at very specific points in the song.  I've come to the conclusion that the difference between mp3 and flac is so small that it would not, in any circumstance, affect my enjoyment of the music.  So I'm no longer bummed :)

post #19 of 23
Quote:
Originally Posted by xnor View Post

Decoding to 64-bit internally is completely pointless with mp3. First of all, the input is usually 16 bits, secondly, most decoders do not even come close to 32 bits (most do around 16 undistorted bits) and thirdly, floating point output usually has to be quantized to 16 or 24 bit integers clipping anything above 0 dBFS in the process.

 

=> Converting to WAV will not affect the sound.

 

The source is 16-bit, the resulting floating-point data is not 16-bit. You should output lossy files at the highest bit-depth your hardware supports.

This whole presentation is interesting, but you can see how encoding to lossy formats such as MP3 requires gain reduction to avoid distortion: http://www.youtube.com/watch?v=BhA7Vy3OPbc#t=9m23s

post #20 of 23
Quote:
Originally Posted by StudioSound View Post

The source is 16-bit, the resulting floating-point data is not 16-bit. You should output lossy files at the highest bit-depth your hardware supports.

This whole presentation is interesting, but you can see how encoding to lossy formats such as MP3 requires gain reduction to avoid distortion: http://www.youtube.com/watch?v=BhA7Vy3OPbc#t=9m23s

It doesn't matter what the output of the decoder is. Even if it is 128 bits you can quantize it to 16 bits if the input was 16 bits. The lowest bit of the input was probably just dither or pure noise anyway.

If you quantize to 24 bits you just have distorted noise in the lower 8 bits.

 

Decoders and audio players, afaik, do not attenuate lossy tracks for the simple reason that the amount of attenuation would have to be guessed and that the peaks are usually rare and do not cause audible problems.

post #21 of 23
Quote:
Originally Posted by xnor View Post


Decoders and audio players, afaik, do not attenuate lossy tracks for the simple reason that the amount of attenuation would have to be guessed and that the peaks are usually rare and do not cause audible problems.

 

Try analyzing your library with the new JRiver 19 beta. You will find that these inter-sample peaks are common, particularly with lossy files.

post #22 of 23

I'm using foobar2000 which has an EBU R128 compliant loudness scanner (used for ReplayGain) included for a couple of months now - I know that lossy files have peaks above 0 dBFS, but they are usually not severe nor all over the place. Of course if you take a track that digitally clips to begin with you can get many such peaks but that doesn't matter anymore in this case, does it?

 

Still, afaik no player is attenuating such files unless you activate something like ReplayGain with clipping protection and have analyzed the files. Peaks will simply be clipped on quantization.


Edited by xnor - 8/28/13 at 2:01pm
post #23 of 23
Quote:
Originally Posted by xnor View Post

I'm using foobar2000 which has an EBU R128 compliant loudness scanner (used for ReplayGain) included for a couple of months now - I know that lossy files have peaks above 0 dBFS, but they are usually not severe nor all over the place. Of course if you take a track that digitally clips to begin with you can get many such peaks but that doesn't matter anymore in this case, does it?

 

Still, afaik no player is attenuating such files unless you activate something like ReplayGain with clipping protection and have analyzed the files. Peaks will simply be clipped on quantization.

 

Does Foobar measure True Peak Level? I have Lossless files which have peaks well above 0 dB. The few lossy files I have go even further above 0 dB.

The problem is that music is often normalized to have sample peaks of -0.1 dBFS. This often leads to inter-sample clipping during playback.

 

JRiver 19 will avoid inter-sample peaks when playing back analyzed files.

New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Sound Science
Head-Fi.org › Forums › Equipment Forums › Sound Science › I'm bummed :(