VBR HR format?

Discussion in 'Computer Audiophile: Software, Configs, Tools' started by Stuff Jones, Nov 15, 2017.

  1. Stuff Jones

    Stuff Jones Friend

    Friend
    Joined:
    Feb 3, 2016
    Likes Received:
    466
    Dislikes Received:
    8
    Trophy Points:
    63
    Back in the day when I ripped all my CDs to MP3s, I always used to rip in VBR 192. This seemed like a zero SQ cost way to save storage space.

    Currently most of my music is 44/16 FLAC and I don't download HR audio files for the same reason (well also cost). All my music is on my laptop where still storage space is an issue (512 GB HDD).

    Is there a variable bitrate high resolution format similar to for MP3 which fluctuates between say 88/24 and 44/16 as per the demands of the music? Is this even theoretically possible?

    Disclaimer: I'm not sure if this is the correct sub-forum.
     
    Last edited: Dec 2, 2017
  2. an0n1mate

    an0n1mate Rando

    Joined:
    Nov 17, 2016
    Likes Received:
    0
    Dislikes Received:
    0
    Trophy Points:
    1
    Location:
    Lima
    I don't undestand the question quite well... if you want variable sample rate, that's not posible (as far as I know). If you are refering to a lossy variable bitrate format that maintains the orignal bit depth of the lossless file, it's doable. I use OGG Vorbis for that purpose, at -q 0.6 it maintains the sample rate of the file (44/48/88.2/96) while being VBR 192. MP3 only goes up to 48000 Hz IIRC.
     
  3. Taverius

    Taverius Smells like sausages

    Friend
    Joined:
    Dec 27, 2017
    Likes Received:
    405
    Dislikes Received:
    1
    Trophy Points:
    63
    Location:
    Rapallo, Italy
    Possibly slight necromancy, but I thought I could clarify on why the question doesn't truly make sense.

    It is theoretically possible, but silly, for a number of reasons:

    1) many many dacs will briefly pause when switching resolution/bitrate. So you'd resample the 44 to 88 on output, and at that point there's better ways to save space in lossy codecs.

    2) 24/16 is meaningless in lossy codecs, they all use floating point internally, and allocate 'resolution' via bitrate.

    3) if your source is 44/16, upsampling and padding to 24 does nothing except make it take more space.

    Bottom line: just use LAME MP3 -V0 for your lossy stuff and quell your nervosa.

    Or use OGG, or Opus, as they can both achieve higher maximum bitrates, and have newer - and better - codecs; but be aware that LAME MP3, old as it may be, has also been tuned over decades to solve thousands of problem samples.

    LAME needs higher bitrates to be 'transparent' - that's an argument all in itself - but it is also remarkably consistent in sound quality, to a level that no AAC, ALAC, OGG, Musepack, Ogg, Opus etc encoder has ever achieved. Something to be said for tried and tested.

    Also, at a certain point with increasing lossy bitrates you get close enough you might as well just use FLAC.

    I could pontificate further but really, HydrogenAudio is the place to discuss audio codecs in depth.
     
  4. Stuff Jones

    Stuff Jones Friend

    Friend
    Joined:
    Feb 3, 2016
    Likes Received:
    466
    Dislikes Received:
    8
    Trophy Points:
    63
    @Taverius Obviously the source would have to be the highest resolution used and then the codec would down sample based lower data requirements in particular passages of music. I obviously don't know much about this so I apologize if I'm not fully understanding your response.
     
  5. Azteca

    Azteca Friend

    Friend
    Joined:
    Sep 26, 2015
    Likes Received:
    1,259
    Dislikes Received:
    1
    Trophy Points:
    93
    Location:
    Cincinnati, Ohio
    I'll do my best to give a response that addresses the misconceptions you might have going in. This is overkill, listen to Taverius and just check out these listening test results if you want to know about good encoders. I say use qAAC unless you have compatibility problems, or use V0 MP3. Opus is great but low compatibility and still being actively improved. Here's a good listening test from HydrogenAudio from a few years ago: http://listening-test.coresv.net/results.htm Note that I wouldn't consider those transparent bitrates. V0 is ~256kbps, I'd use the same VBR for AAC.
    That gives you what to do but not why. Feel free to skip this below but I guess it will be the start of my basic digital explainer article I've been meaning to write as I teach fundamentals of sampling rate, bit depth, psychoacoustics etc. every semester anyway:
    • FLAC is already VBR. WAV is CBR. A 70 minute disc with old mono recordings will be 700MB in WAV. It could be 200 in FLAC. FLAC is lossless, but uses the least data possible to achieve this.
    • A variable sample rate doesn't make sense. To make sense of this, consider the Nyquist theorem on which all digital sampling is based. To accurately capture a given range of frequencies, you must take that frequency (in Hz), double it, and add a bit more. The result is how many samples per second you need. We use 44.1kHz because it goes 20Hz-20kHz (top of human hearing), plus some room for filtering (22kHz). 22*2=44. Add a bit more and it's 44.1kHz. So that's 44,100 samples per second. If you jump up to, say, 88kHz, you need twice as many samples per second, which means more data any way you slice it.
    • An attempt at a lossy psuedo-hires codec is already out there and failing: MQA. You can go read all you want about that if you'd like. I'd recommend Archimago's Musings.
    • Lossy codecs are built on balancing perceptual losses in fidelity with reduction in file size. For this to make sense, they also need to be meaningfully smaller than a lossless file. Lossy codecs work from the accepted psychoacoustics research showing our sensitivity to various frequencies (check out equal loudness countours).
    • In short, extreme highs (~17kHz+) are a very easy place to trim data as they must be far stronger for us to perceive them and we naturally lose them as we age. Anything above the 22kHz that 44.1kHz sampling rate allows is nice to look at on a spectrogram, is perhaps the original format it was recorded in, but lossy formats are never going to be for perfectionists. It's a trade off. So making lossy 88.2 doesn't make much sense.
    • You didn't really ask about bit depth but: 0dB is the absolute bottom of human sensitivity to sound as based on research. The best mastering studio in the world you'd ever be in has a noise floor ~20dB at least. In normal life, ~35-40dB is probably the best you'll encounter. 120dB is big concert being mixed too loud, threshold of pain, you are risking permanent damage. so 120 (top) minus 30 (bottom) = 90dB of dynamic range we want to capture. Well what do you know, 16 bits captures 96dB of range. 1 bit = 6dB of dynamic range. 24 bit is mainly used for studio/production purposes because its massive range (24*6=144dB) allows for conservative gain staging to ensure lots of headroom (extra room up top for loud passages to avoid clipping and distortion). This is great, I use it every day in my work recording classical music. But when it's all said and done, and you have it mastered or just normalize it to peak at 0dB, 16 bit is generally enough. If you aren't a classical nerd, the appeal of a 24-bit download diminishes greatly because the music is way less dynamic and is probably not close to utilizing 16 bits. There are arguments about avoiding dithering (the amplitude equivalent of downsampling/sample rate conversion) , or accurately reproducing the magnetic tape noise floor (a la scanning old films in 4K to preserve the grain). All of this goes out the window when you're talking about file space constraints and listening in noisy environments. They should absolutely capture, mix and master in 24-bit. But 16-bit should be fine for you as a delivery format.
    Or, uh, you could just buy a good external drive for $68 and stop sweating it. That's like three hi-res purchases.
     
    rtaylor76, mrflibble, Vent and 5 others like this.
  6. Azteca

    Azteca Friend

    Friend
    Joined:
    Sep 26, 2015
    Likes Received:
    1,259
    Dislikes Received:
    1
    Trophy Points:
    93
    Location:
    Cincinnati, Ohio
    Also Wavpack exists but its use case is kinda confusing to me. Compatibility is even lower.
    http://www.wavpack.com
     
  7. Taverius

    Taverius Smells like sausages

    Friend
    Joined:
    Dec 27, 2017
    Likes Received:
    405
    Dislikes Received:
    1
    Trophy Points:
    63
    Location:
    Rapallo, Italy
    I'll add a little onto what
    @Azteca wrote:

    One important thing I explained badly because I'm awfulbad at being concise:

    Internally a lossy codec is the mathematical representation of an approximation of the input, like (terrible wrong but illustrative example) a sine wave would just be the three values for frequency, amplitude and phase, as arbitrary precision numbers.

    Inside a lossy audio file, sampling frequency and bit depth have no meaning, although because of reasons, you want to play it back at same sampling rate it was fed, and indeed can't do otherwise.

    So adjusting the bit-depth dynamically is meaningless, and likewise for sampling rate.
     

Share This Page