Experiment: does transcoding from a low bitrate to a high bitrate improve quality?

[Febs]

After reading the off-topic discussion in this thread, I did a very simple experiment to test the hypothesis that the quality of an MP3 can be improved by transcoding to a higher bitrate.

Hypothesis: transcoding an MP3 file from a lower bitrate to a higher bitrate increases the quality of the file.

Methodology:

• Rip a 30 second clip of a song to .wav format.
• Encode from .wav to 64 kbps CBR MP3.
• Transcode from 64 kbps to 96 kbps CBR MP3
• Transcode from 96 kbps to 128 kbps CBR MP3
• Transcode from 128 kbps to 160 kbps CBR MP3
• Transcode from 160 kbps to 192 kbps CBR MP3
• Transcode from 192 kbps to 224 kbps CBR MP3
• Transcode from 224 kbps to 256 kbps CBR MP3

If the hypothesis that converting an MP3 from a lower bitrate to a higher bitrate makes a file sound better is true, then step 3 would result in a higher quality file than step 2.

Likewise, if the hypothesis is true, step 4 would yield a higher quality file than step 3, step 5 would yield a higher quality file than step 4, and so on. This could be expressed as:

8 > 7 > 6 > 5 > 4 > 3 > 2

It follows that if the hypothesis is true, then 8 > 2.


Results:

64 kbps clip
256 kbps clip
Original .wav clip

You can listen to the results and decide for yourself, but to my ear, the 256 kbps file is appreciably worse than the 64 kbps file, and both are appreciably worth than the original .wav. Thus, it is clear to me that transcoding to a higher bitrate does not increase quality.

Notes:

1. The clip I used was XTC's Mayor of Simpleton from Oranges and Lemons. There was no particular reason that I selected this song other than that I happened to have the CD with me.

2. You could achieve the same results by transcoding once from any lower bitrate to any higher bitrate. However, in that scenario, the difference between the two files is not nearly as noticeable, and--as seen in the original thread linked to above--the artifacts introduced by a single transcode can be subjectively interpreted as an improvement to the sound of an mp3. Doing the experiment with 6 separate transcoding steps, however, makes the file degradation obvious. If the hypothesis is correct, and transcoding from a low bitrate to a high bitrate increases sound quality, then it should be correct not only for one transcode, but for each individual transcode in the experiment.

 

[Bunnyears]

Here's my understanding: when you compress files you are actually also removing information in order to facilitate the compression. If you then "re-expand" the files by reencoding at a higher bitrate, you don't really get anything substantially better because you aren't replacing the information that was taken out in the original compression. You are actually taking a small file and re-compressing it, thus making it smaller (take 20% of X and then take another 80% of the 20% and you end up with 16% of X not 100%). No wonder it sounds worse when you try to "re-expand" the files.

 

[Febs]

Absolutely. Each re-compression introduces additional compression artifacts and quantization noise. In the earlier thread that I linked to, however, there was some insistence that transcoding to a higher bitrate could improve quality.

 

[claymon]

simply put, you can't polish a turd.

 

[erdinc]

The first file reads as 176 kbps, not 64. I player with foobar and windows media player.

 

[guynamedjohn]

If reencoding to a higher bit rate could improve sound then why stop at 64kbs? Why not go for 1kbps? Then put the transcoder on the player and I could fit my whole collection on an iPod nano.

Seriously, though, when you compress you lose data. It's a fair to debate whether what you lose is significant or not, but lost it is.

 

[Factor]

This experiment is quite the forehead-slapper.

 

[cire]

oh course it wouldn't. you can't make something out of nothing.

 

[guynamedjohn]

Quote:

Originally Posted by cire oh course it wouldn't. you can't make something out of nothing.

Yes. You can't make chicken salad out of chicken crap.

 

[Febs]

Quote:

Originally Posted by marcelproust The first file reads as 176 kbps, not 64. I player with foobar and windows media player.

Whoops. You're right. Thanks for pointing that out. I used a bad switch. I'll fix that and repost the files.

Edit: files in the original post are now correct bitrates.

Quote:

Originally Posted by Factor This experiment is quite the forehead-slapper.

I know, it seems obvious, but there are those who insist that transcoding to a higher bitrate increases quality, and I wanted to put together a demonstration that it just isn't true.

 

[Bunnyears]

Quote:

Originally Posted by Febs I know, it seems obvious, but there are those who insist that transcoding to a higher bitrate increases quality, and I wanted to put together a demonstration that it just isn't true.

I was wondering why you were going to all of this trouble!

 

[titaniumx3]

Quote:

Originally Posted by Febs I know, it seems obvious, but there are those who insist that transcoding to a higher bitrate increases quality, and I wanted to put together a demonstration that it just isn't true.

Obviously encoding from lower bitrate to higher bitrate won't increase quality for reasons already discussed, but percieved sound quality is very subjective and doesn't always correlate with the objective sound quality.

I've seen many tests done where people actually preffered 128kbps mp3s to the original. The mp3 compression artifacts may have added a smoothening effect that suited harsher material. Therefore increasing bitrate from lower bitrate may add more "beneficial" compression artifacts leading to the same effect.

 

[Febs]

I recognize that, and I wrote something very much to that effect in my original post:

Quote:

However, in that scenario, the difference between the two files is not nearly as noticeable, and--as seen in the original thread linked to above--the artifacts introduced by a single transcode can be subjectively interpreted as an improvement to the sound of an mp3. Doing the experiment with 6 separate transcoding steps, however, makes the file degradation obvious.

 

[jagorev]

Heh. Nicely done. That 256 clip was AWFUL.

 

[jbloudg20]

common sense