3 ms·

> That said, there are people who can hear the difference between 320kbps mp3s and uncompressed, because they can sense the faint temporal errors that can make

> That said, there are people who can hear the difference between 320kbps mp3s and uncompressed, because they can sense the faint temporal errors that can make a sound a touch before the sound they should follow. You got a link to some independent blind listening test results showing someone has this ability? I’m extremely skeptical of the above claim.

bayindirh 4y ago
You can try it yourself: http://abx.digitalfeed.net/ http://abx.digitalfeed.net/ Moreover, there's a file which I just made [0] which shows the sound difference between a FLAC and a MP3 file. How I made this? 1. Rip the CD to FLAC. 2. Convert FLAC to MP3 with lame (320CBR, 44.1KHz, Compression level 5). 3. Load both to Audacity, subtract MP3 from FLAC. 4. Amplify remainder to 0.0dB. 5. Render to FLAC. So, yes, there's audible difference. P.S.: To verify I tried this process with two copies of the same FLAC file, and what remains from subtraction is a clear 0. So, the process is working. [0]: https://share.getcloudapp.com/5zurmeLz https://share.getcloudapp.com/5zurmeLz [0]: http://0x0.st/oBAz.flac http://0x0.st/oBAz.flac (alternative link if the first one dies).
- roelschroeven 4y ago
  You found a difference between the same audio compressed losslessly and lossy. That's not a surprise; it would be a surprise if there weren't a difference. The whole point of lossy compression is that it throws away some information to get better compression rates. Of course it's not going to be identical. But that doesn't prove anything, and more specifically it doesn't prove that there is an audible difference. What matters is the perception of the sound. In the end, the only reliable differentiator is a properly set up double-blind listening test.
  - bayindirh 4y ago
    > That's not a surprise; it would be a surprise if there weren't a difference. The surprise part is not the difference itself, but the amount and legibility of difference even with a state of the art compressor at maxed-out settings. > But that doesn't prove anything, and more specifically it doesn't prove that there is an audible difference. On the contrary, it proves that some audible information is stripped from an MP3 file, even at highest settings, and this stripped detail inevitably adds more information to the sound stage. It wouldn't be there if there was no contribution. Therefore, audibility of this information depends on the audio chain used. As I said before, a semi-decent amp with a good pair of speakers can render this, and the difference is not Oh, I hear a cymbal here, but Oh, this recording is really enjoyable and immersive. The equipment I cite here is not a 5-figure system either. A good vintage amp (Akai, Sony, Marantz, Technics, etc.) with an "affordable" high end speaker (Heco Celan GT/Revolution 302, which are smallest members of their respective families) is more than enough for hearing the difference. Investment in audio equipment has decent return in terms of quality up to a certain point. Past that point, it becomes meaningless for most people and scenarios. Also, music has layers, and one listens a song completely in "multiple passes", this is why ABX tests play the same song 5-10 times and ask for evaluation. Moreover, as per the study published in Audio Engineering Society in 2016 [0], MP3 changes the emotional perception of the songs, again verifying there's audible differences. [0]: (Open Access) https://www.aes.org/tmpFiles/elib/20220528/18523.pdf https://www.aes.org/tmpFiles/elib/20220528/18523.pdf
    - roelschroeven 4y ago
      > On the contrary, it proves that some audible information is stripped from an MP3 file, even at highest settings, and this stripped detail inevitably adds more information to the sound stage. It wouldn't be there if there was no contribution. It doesn't prove that at all. Our perception of sound is quite complex. Sounds that would be very noticeable in isolation can become inaudible in combination with other simultaneous sounds, or even sounds that came shortly before. MP3 and other lossy compression schemes use a psychoacoustic model that takes all kinds of things like that into account. A - B being audible does not mean that A and B can be audibly differentiated. Perception of sound is not maths. > Moreover, as per the study published in Audio Engineering Society in 2016 [0], MP3 changes the emotional perception of the songs, again verifying there's audible differences Yes, as I said, listening tests are what we should use as datapoints. Not mathematical differences between the waveforms. That being said, in this case I see the study uses MP3s with bitrates of 32, 56 and 112 kbps. Is that a serious study? Those are very low bitrates for MP3! The authors of the study picked the lowest bitrate nearly indistinguishable from the original sound (i.e. already not perfect by their own admission), and then picked two bitrates even lower. No wonder the test subjects are going to hear a difference. Nobody is going to argue that those MP3 bitrates should be considered good quality. Any result obtained at these low bitrates is completely irrelevant for compression with even moderately decent bitrates (let's say starting from 160 with a decent encoder such as LAME). I don't really care one way or another whether people can hear a difference between lossy compressed audio and the original (for me, with my ears and on my equipment, 192 kpbs MP3s are already more than good enough). What I do care about is the use of proper methodology.

bayindirh 4y ago

You can try it yourself: http://abx.digitalfeed.net/ http://abx.digitalfeed.net/ Moreover, there's a file which I just made [0] which shows the sound difference between a FLAC and a MP3 file. How I made this? 1. Rip the CD to FLAC. 2. Convert FLAC to MP3 with lame (320CBR, 44.1KHz, Compression level 5). 3. Load both to Audacity, subtract MP3 from FLAC. 4. Amplify remainder to 0.0dB. 5. Render to FLAC. So, yes, there's audible difference. P.S.: To verify I tried this process with two copies of the same FLAC file, and what remains from subtraction is a clear 0. So, the process is working. [0]: https://share.getcloudapp.com/5zurmeLz https://share.getcloudapp.com/5zurmeLz [0]: http://0x0.st/oBAz.flac http://0x0.st/oBAz.flac (alternative link if the first one dies).

roelschroeven 4y ago
You found a difference between the same audio compressed losslessly and lossy. That's not a surprise; it would be a surprise if there weren't a difference. The whole point of lossy compression is that it throws away some information to get better compression rates. Of course it's not going to be identical. But that doesn't prove anything, and more specifically it doesn't prove that there is an audible difference. What matters is the perception of the sound. In the end, the only reliable differentiator is a properly set up double-blind listening test.
- bayindirh 4y ago
  > That's not a surprise; it would be a surprise if there weren't a difference. The surprise part is not the difference itself, but the amount and legibility of difference even with a state of the art compressor at maxed-out settings. > But that doesn't prove anything, and more specifically it doesn't prove that there is an audible difference. On the contrary, it proves that some audible information is stripped from an MP3 file, even at highest settings, and this stripped detail inevitably adds more information to the sound stage. It wouldn't be there if there was no contribution. Therefore, audibility of this information depends on the audio chain used. As I said before, a semi-decent amp with a good pair of speakers can render this, and the difference is not Oh, I hear a cymbal here, but Oh, this recording is really enjoyable and immersive. The equipment I cite here is not a 5-figure system either. A good vintage amp (Akai, Sony, Marantz, Technics, etc.) with an "affordable" high end speaker (Heco Celan GT/Revolution 302, which are smallest members of their respective families) is more than enough for hearing the difference. Investment in audio equipment has decent return in terms of quality up to a certain point. Past that point, it becomes meaningless for most people and scenarios. Also, music has layers, and one listens a song completely in "multiple passes", this is why ABX tests play the same song 5-10 times and ask for evaluation. Moreover, as per the study published in Audio Engineering Society in 2016 [0], MP3 changes the emotional perception of the songs, again verifying there's audible differences. [0]: (Open Access) https://www.aes.org/tmpFiles/elib/20220528/18523.pdf https://www.aes.org/tmpFiles/elib/20220528/18523.pdf
  - roelschroeven 4y ago
    > On the contrary, it proves that some audible information is stripped from an MP3 file, even at highest settings, and this stripped detail inevitably adds more information to the sound stage. It wouldn't be there if there was no contribution. It doesn't prove that at all. Our perception of sound is quite complex. Sounds that would be very noticeable in isolation can become inaudible in combination with other simultaneous sounds, or even sounds that came shortly before. MP3 and other lossy compression schemes use a psychoacoustic model that takes all kinds of things like that into account. A - B being audible does not mean that A and B can be audibly differentiated. Perception of sound is not maths. > Moreover, as per the study published in Audio Engineering Society in 2016 [0], MP3 changes the emotional perception of the songs, again verifying there's audible differences Yes, as I said, listening tests are what we should use as datapoints. Not mathematical differences between the waveforms. That being said, in this case I see the study uses MP3s with bitrates of 32, 56 and 112 kbps. Is that a serious study? Those are very low bitrates for MP3! The authors of the study picked the lowest bitrate nearly indistinguishable from the original sound (i.e. already not perfect by their own admission), and then picked two bitrates even lower. No wonder the test subjects are going to hear a difference. Nobody is going to argue that those MP3 bitrates should be considered good quality. Any result obtained at these low bitrates is completely irrelevant for compression with even moderately decent bitrates (let's say starting from 160 with a decent encoder such as LAME). I don't really care one way or another whether people can hear a difference between lossy compressed audio and the original (for me, with my ears and on my equipment, 192 kpbs MP3s are already more than good enough). What I do care about is the use of proper methodology.