FLAC 1.4 Released With AArch64 Optimizations, Faster x86_64 FMA

[Brisse]

Anux Another thing which impressed me when I experimented with Opus back in the days when it was fairly new was it's resistance to generation loss. AAC and Vorbis are pretty great when encoding music at decent bitrate (~160kbit or more) from the master or a lossless copy of the master, but when encoding from a lossy source then quality quickly degrades. In my experiments, AAC and Vorbis ended up a garbled mess after some 4-6 encoding passes. Basically, the compression artifacts are pretty much the same each pass which means they amplify themselves every time. Opus uses some trickery to add entropy to the compression artifacts so they don't end up amplifying as much during re-encoding, and I was able to re-encode a music test file more than a dozen times without all that much quality degradation. Quite impressive.

I still stick to Vorbis for my music library though, mostly because it's tried and tested, I always convert from a lossless source, and also because Opus doesn't support 44.1Khz. The latter is not a huge deal with high quality resampling easily available, and even built into the encoder itself for convenience, but it is a compromise they made due to their main focus being on web communication rather than music and such.

[Anux]

Anux
I still stick to Vorbis for my music library though, mostly because it's tried and tested, I always convert from a lossless source, and also because Opus doesn't support 44.1Khz. The latter is not a huge deal with high quality resampling easily available, and even built into the encoder itself for convenience, but it is a compromise they made due to their main focus being on web communication rather than music and such.

They actually "support" every frequency <= 48 kHz because of the way the wavelet encoding works. The upsampling is not lossless but neither is opus and the upsampling is totally transparent. For a lossless codec that would have been a no go but in lossy every trick that you can't notice is allowed.

[willmore]

1.4.1 is coming within the next week. There were some build errors on certain platforms.

[willmore]

And now there's some discussion about build issues on OSX, so the next point release to clean this up may be pushed back.

[arQon]

I disagree with your characterization of MP3 as an adequate format for finicky end users; even in old age, my own ears (worn out with lots of pro music-making) can tell the difference between MP3 and versus Opus 'lossy' conversion at their highest bit rates. For my own use (even in the car, where my audio files must be subjected to a TON of compression) MP3 @ 320 kb sounds lifeless, compared to FLAC. My auto head unit doesn't understand OPUS.

It's the encoder that matters really, not the format. A 192K MP3 created with "modern" LAME has significantly better fidelity than a 320K one created two decades ago. It's not really credible that you wouldn't find modern 320K indistinguishable from WAV in a blind test - because literally nobody does; including all the people who imagine they do, once that external knowledge is removed.

There are a lot of aspects to Opus that are much better designed than MP3 (notably, the ability to downsample without needing to transcode), certainly, but as far as sound quality per bitrate goes the gap has shrunk enormously in the last 15 years, especially at higher bitrates.
Which is good, since, as you say, almost nothing supports Opus, and now that MP3 no longer has any patent encumbrance that fight is even more lost than it already was.

These days, I keep all my audio in FLAC, and transcode it depending on the player's capabilities and capacity. For the PC I just listen to the FLAC masters, but for both the car and the phone that means 320K MP3 these days, because that's an excessively high bitrate but it all fits on the SD cards with some room to spare, so I don't have to nickel and dime it the way I did when I first ripped all my CDs to ogg a million years ago. Much though I appreciated it then, there's no justification for ever using ogg any more unless you're *absurdly* short on storage space - in which case the better solution by far, if possible, is to just buy more storage, since using the better format only improves the sound quality from "terrible" to "poor".

[rickst29]

It's the encoder that matters really, not the format. A 192K MP3 created with "modern" LAME has significantly better fidelity than a 320K one created two decades ago. It's not really credible that you wouldn't find modern 320K indistinguishable from WAV in a blind test - because literally nobody does; including all the people who imagine they do, once that external knowledge is removed.

<snip>

These days, I keep all my audio in FLAC, and transcode it depending on the player's capabilities and capacity. For the PC I just listen to the FLAC masters, but for both the car and the phone that means 320K MP3 these days, because that's an excessively high bitrate but it all fits on the SD cards with some room to spare, so I don't have to nickel and dime it the way I did when I first ripped all my CDs to ogg a million years ago. Much though I appreciated it then, there's no justification for ever using ogg any more unless you're *absurdly* short on storage space - in which case the better solution by far, if possible, is to just buy more storage, since using the better format only improves the sound quality from "terrible" to "poor".

I won't disagree with respect to playback in cars and phones in open air. A few wired headphones may show the difference when played from a phone, but most bluetooth in-ear phones and foam-wrapped over the ear phones will also compromise reproduction badly. In response to your post, I just did a blinded test on my home set, with my wife choosing between FLAC and 320K "insane" MP3 versions of a Classical recording I know very well. (She was "choosing" with the computer mouse and solid mouse pad from behind my listening location.) Starting from the beginning and ending at the same spot. I was 100% correct in identifying 4 playbacks of MP3, and 75% correct in identifying 4 playbacks of FLAC (labeling one FLAC playback as MP3). That's 87% overall correct in identifying the compromised versus original format correctly.

In the FLAC version, the violin section is presented with realistic 'airiness', and the texture of flute and oboe (especially oboe) is also better. Two low-volume tympani hits are also superior in the FLAC version. My encoder 'libmp3lame0' is version 3.1 plus a couple of more recent fixes (not audio related). My libmpg123.0 player is at version 1.30.2 (and possibly newer than yours). My libmad0 is barely a week old and already contains the not-audio-decode-related fix for CVE-2017-8373.

[arQon]

most bluetooth in-ear phones and foam-wrapped over the ear phones will also compromise reproduction badly.

Thanks for starting my day with the cringe-inducing mention of BT and audio in the same sentence... :P
No question, those are both terrible, no matter how many EQs Apple throws into the stream.

> That's 87% overall correct in identifying the compromised versus original format correctly.

Then either your hearing is beyond exceptional - which as you know is highly unlikely given either your age or the years spent in a studio, let alone both - or Something Is Wrong, but I have no idea what. It's not like the studies aren't out there, and since they're generally university-based they have both an ample pool of young healthy ears to work with, as well as the clout/money to pull in "experts", and I haven't seen one in at least 10 years where a single individual - not just the group average - was statistically better than pure guesswork.

The pieces you call out, and the description, does of course match what you'd expect from a lossy encoder not having the bits to do a good job, but that shouldn't be happening even if e.g. the tympani is concurrent with a piccolo etc.

Also, congratulations on your choice of wife.