Announcement

Collapse
No announcement yet.

Opus 1.3 Released - One Of The Leading Lossy Open-Source Audio Codecs

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #41
    Originally posted by Weasel View Post
    It's simple logic, really. You don't need any expertise to prove this. If an encoder decides that it's worth it to encode frequencies above 20k, it must mean it has done a perfect job (acoustically) on lower frequencies. Why else would it start encoding freqs above 20k in AAC?
    Nope. Not that simple. If you target a certain bitrate, and spend more bits on high frequencies, you must sacrifice bits otherwise used for lower frequencies to reproduce those higher frequencies.

    "If an encoder decides that it's worth it"
    Encoders aren't perfect. These things depend on tunable parameters that the encoder developers set for you based on their testing. Sometimes they make mistakes, or they tune it in a way which fits a certain audio sample well but there might be another audio sample where it doesn't work well. Some encoders (such as reference Vorbis) provides flags for the user to override default parameters, although one normally would not do so.

    Comment


    • #42
      Originally posted by Brisse View Post
      Nope. Not that simple. If you target a certain bitrate, and spend more bits on high frequencies, you must sacrifice bits otherwise used for lower frequencies to reproduce those higher frequencies.
      So what happens if there's no information that matters to encode in lower bits anymore?!?? Maybe you encoded literally all frequencies of the signal and with a negligible quantized error on phases, which I bet is what most decent encoders actually do. That's why they start encoding higher frequencies. Have you ever subtracted (after shifting) the two signals to see just how faint the differences are?

      You guys have this misconception that you can't encode the entire information except if it's uncompressed (or at least losslessly compressed) but that is NOT true. Lossy compressors all encode information in the FREQUENCY domain, not PCM like lossless codecs.

      In the frequency domain, the information is stored differently. A simple sine wave with an exact frequency bin bandwidth is encoded with very little data in the frequency domain, and yet you need a lot of data in the normal time domain (PCM).

      So let's ask a theoretical example: what happens if your signal has some simple sine waves and then a lot of information above 16k? And you encode it at 256kbps? The lower spectrum requires very little data so ofc the codec will start encoding higher frequencies with no loss.

      BTW you can take out the high frequencies yourself, if you want, with an audio editor or spectrum editor. The bitrate will go down which means less information overall because the encoder WILL NOT put "more bits" in the lower spectrum since it finds it either useless or pointless (if phase is quantized already to the minimum, there's nothing else it can add, literally).

      Originally posted by Brisse View Post
      Encoders aren't perfect.
      They aren't, but they are created by humans.
      Opus is not perfect and it's also created by humans.

      You're basically claiming that the Opus devs are smarter than literally everyone else who makes an encoder because they forcefully take out freqs above 20k no matter the signal.
      Last edited by Weasel; 11 November 2018, 08:50 AM.

      Comment


      • #43
        Originally posted by SciK View Post
        Such trust is beautiful, but it seems a little naïve.

        Let me reiterate my previous question: if we assume that it has indeed done a perfect job on audible frequencies, why bother at all with the higher frequencies and create a bigger file as a result?
        Because you asked it for that bitrate or "quality" (the latter depends on encoder and is subjective).

        I don't know how to tell you this, but you don't tell other people what sort of information they want to encode. If they want to keep what you consider inaudible, then they should be able to by ramping up the bitrate.

        Consider the fact that many people even buy 96khz "high quality" lossless files despite the fact it's beyond useless.

        Furthermore, if it DID drop the high frequencies, it would encode with less bitrate than Opus (this actually happens btw, if you remove those frequencies yourself in a spectral editor), proving that it is superior to Opus as it needs less information to encode the same perceived quality.

        Comment


        • #44
          Originally posted by Weasel View Post
          Consider the fact that many people even buy 96khz "high quality" lossless files despite the fact it's beyond useless.
          Yeah… that’s their problem, I would say. I already don’t understand them, and wanting to lossily encode such high frequencies goes even further beyond my comprehension.

          Originally posted by Weasel View Post
          Furthermore, if it DID drop the high frequencies, it would encode with less bitrate than Opus (this actually happens btw, if you remove those frequencies yourself in a spectral editor), proving that it is superior to Opus as it needs less information to encode the same perceived quality.
          Again, that doesn’t prove much, unless you have tried asking Opus for that same bitrate and compared the result aurally. (Also, as far as I know, Opus doesn’t yet have a quality-driven encoding mode, so what do you mean by encoding with less bitrate than Opus?)
          Last edited by SciK; 13 November 2018, 04:49 PM.

          Comment


          • #45
            Originally posted by SciK View Post
            Again, that doesn’t prove much, unless you have tried asking Opus for that same bitrate and compared the result aurally. (Also, as far as I know, Opus doesn’t yet have a quality-driven encoding mode, so what do you mean by encoding with less bitrate than Opus?)
            I'm saying that AAC encoders will lower the bitrate if the source signal has less information (e.g. missing high frequencies above 20k) instead of "improving the quality" of frequencies below 20k as people think. It probably can't be improved anymore.

            Because it thinks it's doing a perfect job there, and with a diff, it kinda shows it does. Remember that lossy compressors can't encode something perfectly since they use different encoding representation, frequency domain with SFFT windows. Sometimes you can't just throw more bits at it.

            Comment

            Working...
            X