Why does MP3 transcoding lose so much quality?

Topic: Why does MP3 transcoding lose so much quality? (Read 1036 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.

Why does MP3 transcoding lose so much quality?

2020-01-07 07:38:34

Hi all,

I have a number of MP3 files that I want to trim, but can't, for one reason or another. Some are Lame VBR, which often makes trimming leading silence impossible (using mpTrim) without mangling the waveform. Others have noise too close to the audio. I decided to start with one of the latter, a 320k Lame encoding, and to test the transcoding process by converting it to WAV, converting the WAV to a new 320k Lame MP3, converting the new MP3 to WAV and then subtracting one WAV from the other. The two files are exactly the same size, the original encoding process theoretically has already discarded whatever frequencies it needed to, and the output is the same bitrate as the source. I expected the difference file to be inaudible, or nearly so, but it was nearly as loud as the original. I then used the same Lame version (3.93) that was used to encode the original file. That was significantly better, but the difference audio was only about 34 dB down (RMS) from the original -- still very audible. I'm looking for someone who knows enough about the technical details of MP3 encoding to explain why a naive transcoding seems to lose so much of the original sound and what, if anything, can be done to reduce the loss. Thanks in advance.

Re: Why does MP3 transcoding lose so much quality?

Reply #1 – 2020-01-07 13:34:23

Have you listened to the transcoded file? It's a situation I try to avoid, but my experience LAME>LAME when the original file is 320 is that the result is audibly fine. If you want to avoid any transcoding loss, convert the original file to FLAC, do your editing, and save it as FLAC. I've used mpTrim in the past to do what you want (cut gaps from live bootlegs) and it worked pretty well. Do you have the paid version? IIRC, registration unlocked some useful functionality. Another "lossless" MP3 editor you might try is mp3DirectCut.

Re: Why does MP3 transcoding lose so much quality?

Reply #2 – 2020-01-07 14:49:38

ffmpeg also can cut many formats (and mp3 too) without transcoding if you use the right parameters, and it's free and works on any platform.

also please note that
1: listening to the difference in isolation won't tell you whether it's possible to hear the difference together with the signal.
2: lossy compression is not about simply discarding some frequencies, if it was so simple then the files would be much larger.

Re: Why does MP3 transcoding lose so much quality?

Reply #3 – 2020-01-08 18:13:52

Quote

The two files are exactly the same size, the original encoding process theoretically has already discarded whatever frequencies it needed to, and the output is the same bitrate as the source. I expected the difference file to be inaudible, or nearly so

Just FYI, MP4/AAC is MUCH better than MP3 with multiple generations of lossy compression. (But, that probably doesn't apply when transcoding between MP3 and MP4.)

Quote

, converting the new MP3 to WAV and then subtracting one WAV from the other...
... but the difference audio was only about 34 dB down (RMS) from the original -- still very audible

Yes, it's true that if you subtract and you get dead-silence that proves there is no difference in sound. But if subtraction doesn't give you silence, the "difference in sound" is NOT the same as "the sound of the difference".

If you want to demonstrate that for yourself, make a copy of a WAV file and add a little delay to the copy by adding several milliseconds of silence to the beginning. Of course there will be no difference in sound, but if you subtract you'll get a VERY LOUD difference file.

Another experiment you can do is record yourself saying "hello" twice and subtract. (It's important to make two different recordings, but if you wish you can try to match the sound & timing as closely as possible.) Or, record two different people saying "hello", and subtract.

MP3 does actually add a delay, and you can eliminate that by time-aligning before subtracting but the main point of that experiment is to demonstrate that the sound of the difference is not the same as the difference in sound and to show that subtraction can give very misleading results..

Notice