I need YOUR help.

Discussion in 'Discussion & Q&A' started by MarkeyJester, Apr 29, 2018.

?

Which recording sounds better?

  1. Test Sample A

  2. Test Sample B

  3. I cannot tell...

Results are only viewable after voting.
  1. MarkeyJester

    MarkeyJester !%#@ Member

    Joined:
    Jun 27, 2009
    Messages:
    2,645
    Here are two recordings of a sample:

    Test Sample A (.wav)
    Test Sample B (.wav)

    (Here are .ogg files for those of you who cannot download the .wav raw uncompressed versions, though it's recommended you listen to the .wav versions if you can):

    Test Sample A (.ogg)
    Test Sample B (.ogg)

    What I want you to do, is listen to them both, and then vote on which one you think sounds better. Once I have enough votes, I will explain in detail exactly what's significant about the two samples, and why I could not decide between them.

    For those that are clever, please refrain from looking at the waveform visually in a tool, until after you have voted, what you see may psychologically affect your decision, I am only looking for your vote based on what you hear, not what you see.
     
  2. EMK-20218

    EMK-20218 They call me as 'Eduardo Knuckles' Member

    Joined:
    Aug 8, 2008
    Messages:
    933
    Location:
    SEGA Land
    I was able to identify the instruments better in the test B. It sounds way cleaner, or at least it's being easier to diferentiate the instruments for some reason I wasn't able in the test A, such as hi-hats and the bass. I may be wrong, but the sample in test B seems better understandable (yeah, cleaner) while test A is slightly noisy than the sample in test B.
     
    TheInvisibleSun likes this.
  3. Novedicus

    Novedicus Well-Known Member Member

    Joined:
    Aug 26, 2013
    Messages:
    823
    Initially, I couldn't tell the difference, but after listening to both a few more times, I noticed B sounded just a tad cleaner than A, at least to me.

    Disregard my vote on here, as I don't think I can change it once it's set.
     
  4. TheInvisibleSun

    TheInvisibleSun Visible Member

    Joined:
    Jul 2, 2013
    Messages:
    270
    Location:
    Buffalo, NY, USA
    +1 to the "Option B is cleaner sounding" opinion. Since I prefer that, I selected it as 'better', although others might like that some of the instruments are more subdued.
     
  5. nineko

    nineko I am the Holy Cat Member

    Joined:
    Mar 24, 2008
    Messages:
    1,740
    Location:
    italy
    Despite my somewhat trained ears, I couldn't find major differences even after several listenings, so I voted for "I cannot tell". I had the impression that A had a bit more noise, though, but I dismissed it thinking that my mind was playing tricks. Also, both of the samples are quite quiet (did I really type that?).

    After my vote, I looked at the files in a wave editor, and...
    Indeed, since A is capped to a peak of 0.125, it's obvious that the signal-to-noise ratio is worse. My ears were telling the truth, but the difference was so minor that I dismissed it, as I said. Also, the RMS of both samples is indeed low (around 0.041 for both), which explains the quietness as well. The low RMS (in both) and the capped peak at exactly 1/8 (in A) make me assume that you were afraid of clipping, but unless there are technical reasons preventing it, you have a lot of room to raise the volume.

    Also, the tempo is different. At first I thought that you trimmed the files at different times by hand, but it becomes apparent if you try to play them at the same time. I'm not sure how the different tempo is involved into this, though.
     
  6. MarkeyJester

    MarkeyJester !%#@ Member

    Joined:
    Jun 27, 2009
    Messages:
    2,645
    Alright, so the votes are pretty much unanimous at this point, Test Sample B sounds better than Test Sample A. I will explain what the difference is between the two, and why I had trouble deciding (for the record, I thought that B sounded better, but due to circumstances which I will explain in a moment, could not pick that willy nilly).

    But first, just to clear up a few things...
    The reason it's so low, is because of Kega's output amount (I had overdrive turned off by the way):

    [​IMG]
    This is completely beyond my control, and is irrelevant to the test I'm afraid. I would've done a hardware recording, but the audio from my line in is also quiet. Likewise I could not normalise the audio, otherwise it's an unfair test. It should be noted that the FM playback is always louder than DAC (probably due to the necessary fusing of up to 4 operators, and needing more bit range to occupy), this is why many sound drivers set the FM volume to lower than normal, so the volume is in sync with the DAC and PSG.
    That's just dumb luck on my part, I didn't have time to carefully finetune the converter tool to get the exact correct frequency, and went for a near "close enough" rate, not something that I feel would effect the results of the test.

    --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- ---

    Alright, so in the original version of Dual PCM that was released, the samples were converted to 7-bit PCM, this was to ensure that when the samples are fused together, they do not overflow to 9-bit (two 8-bit samples can overflow to 9-bit, two 7-bit samples cannot). The trouble was, the samples were simply too quiet, and lowering the FM would result in hearing a quirk which occurs on model 1's when the total level is low enough (often called the "ladder effect").

    So, with a new volume mechanism in place, we can allow the samples to be 8-bit in the ROM, and instead of converting them to 7-bit, we keep them as 8-bit, but cap them if they go over the 7-bit range, thus ensuring that two samples will still not overflow to 9-bit, and loud enough sounds will cap but the results are "fair", though you do get quantisation noise.

    I wanted to move to a proper capping approach, one where the samples fused together are both 8-bit, and if it overflows to 9-bit, it would cap. This does however cost CPU time to have instructions necessary to ensure it's pushed back to 8-bit, but it does remove the quantisation noise. This comes at the cost of CPU time, and therefore, comes at the cost of sample rate.

    Test Sample A is the fake capped version, one that caps at 7-bits, but runs at around 14600Hz.
    Test Sample B is the real capped version, one that caps on overflow of 9-bits, but runs at around 13500Hz.

    So the question is; which do you sacrifice? Bit rate, or sample rate? Both will have negative effects, but it's which one in the ears of the public is considered better, and should be pursued.

    Thank you all for your response, this helps settle the conflict indefinitely.