Hacker News Reader

Comment by joobus | Hacker News Reader

FLAC is useful for listening to high quality music which hasn't been filtered and compressed. It sounds significantly better than lossy audio. Iscit so hard to comprehend that as more bandwidth comes online all the time, people won't want to compromise on sound quality the way they were forced to in the past?

View on HN

> It sounds significantly better than lossy audio.

Lossy audio can be completely transparent. If you disagree, you need to provide some objective evidence, because all objective evidence points in favour of good lossy compression being indistinguishable from lossless at sensible bitrates.

Look, I've no problem with LAME being supported, but don't make it out to be "significant" in terms of sound quality. Try AAC at 256kbps (what iTunes Store has used for the past decade) or better yet, Opus. Do a serious blinded test and amaze yourself with the results.

It all depends on how sensitive you are to these things, I've done such blind tests, and to my ears the fall in quality was significant, even painful.

Blind tests comparing lossless against what codec at what bitrate? How many runs and how often were you able to correctly identify lossless?

Why to bother with various tradeoffs if you can get the best?(i.e lossless). You can make the same argument on video technologies(i.e HDR). The truth is that people usually adapt so "painful" audio or video quality stops being painful after a while.

"Better" and "best" imply a detectable difference, and the point of a transparent lossy codec is that no such difference should exist.

If it was even remotely painful, it would sail through an ABX test.

It may sail through months long ABX test.

Artefacts that can only be discerned after months of sustained listening? Sounds like you're clutching at straws.

Not an artifact, but loss of variance in music. Same patterns are activating same neurons, so they are tired.

Firstly, on a personal note, I must say getting heavily downvoted on a factual statement and not an opinion is a new and confusing experience for me.

My comparisons were made several years ago using LAME 320 CBR, LAME VBR, OGG, lossless (WAV) 16bit 44.1kHz and lossless (WAV) 16bit 48kHz. I believe at least most professional musicians and sound engineers would be able to identify the difference between all of these; while they might not always be "worse", their sonic character is certainly different.

FWIW I didn't downvote you and I don't think you should be downvoted, but as sjwright points out, if you did indeed make all of these blind comparisons and were able to reliably spot the difference between LAME 320 CBR and lossless and between 44k and 48k, and were able to hear the difference so clearly that you would find the lossy formats jarring or painful to listen to, then given all published research on this topic you would have to have superhuman hearing as most professional musicians and sound engineers in fact can not... so to put it very bluntly there's a concern that you might be selling bullshit.

Exactly. Given how much we know about psychoacoustics, and considering how often people who claim to hear dramatic differences end up failing the most basic ABX tests, it's just not a credible assertion without robust supporting evidence.

ABX tests are putting sensors and short-term memory under stress, while users are reporting about their long-term feelings about music. It's like trying to spot difference between fruits of same kind while consumers have problem with lack of a nutrient.

Audiophiles believe a lot of things with no evidence. How else could you sell them cables costing thousands of dollars? Reminds me of homeopathic medicine. The only reasonable use case for FLAC is archiving music you expect may need to be reencoded later.

I confirm that I feel myself tired when listening low-quality audio even when I cannot spot any audible problems with audio. It's bland.

I.e. I will definitely fail ABX test, but I still feel difference. IMHO, ABX tests are wrong tests for that problem.

If you will fail an ABX test then you don't feel a difference due to an actual difference in quality almost by definition. Maybe you feel a difference because outside an ABX test you know when you are playing lossless files? That's fine but it's not the same thing.

IMHO, lossy codec just removes some variation from music, so some nerves in ear are activated more often than other, so they are tired.

If you had done such blind tests, you would have reached a different conclusion. Or, in the alternative, your hearing is orders of magnitude "better" than anyone else ever tested.

I think you're confusing lossless vs lossy with 192khz vs 44.1khz - in that case I agree with you, no percievable difference.

I can also hear differences between lossless/lossy encodings. If I tweak encoding parameters until I don't notice a difference, there is usually no real space saving afterwards - so why not go with lossess instead?

I'm not confusing these matters. I'm specifically talking about lossless vs lossy. The key point is I'm taking about nominally transparent lossy which for codecs like AAC and Opus is around 160-256 kbps depending on the listener. Even at 256 kbps, that's anywhere between half and quarter the bitrate of FLAC. (Which has an average bitrate around 700 kbps but can swing wildly up or down depending on the specific material.)

I've done ABX tests multiple times, and the vast majority of the time AAC at 200+kbps was indistinguishable from the source (CD WAV), and even 128kbps was fine for a lot of tracks. That said, one or two tracks tripped up the encoder and could reliably be distinguished even at 320kbps (IIRC 20 trials with 100% success rate). Not enough that I'd notice in casual listening, but the difference was audible.

You also need to use a decent encoder - I compressed a video using Handbrake with a 256kbps AAC audio track a few months ago (faac/Windows), and noticed immediately that the audio was bad (I initially assumed the source was bad, but the FLAC track in the source sounded fine). Replaced it with a 192kbps AAC track using Quicktime/OSX and the quality was significantly better.

Yes, ALL lossy codecs have some "killer samples" -- the source material always matters.

It's a sad fact that there are so many sub-par AAC encoders around. Best to use either Apples implementation, or either of those from Fraunhofer -- FhgAAC (free version distributed with Winamp), or the Open Source FDK-AAC that was developed for Android.

opusenc is quite decent, and not patent encumbered unlike AAC encoders.

Yes, Opus is fantastic -- been following it's development since the start due to Telecoms background.

For those of us that care and have software that plays it, it's great but AAC is still the next best thing for compatibility on a wide range of devices (whilst still beating MP3 for sound quality and file size) which makes ad-hoc sharing with non-techies possible.

Hopefully that will continue to improve as it grows in popularity.

For sharing, it's better to use FLAC originals anyway. For playback - Opus is supported by Rockobx, which is useful on Sansa players if you need something very portable (that's what I use sometimes), and is supported on Android too. On desktop systems it's not an issue. Any decent cross platform player can play Opus.

> FLAC is useful for listening to high quality music which hasn't been filtered and compressed. It sounds significantly better than lossy audio.

Doubtful. See https://people.xiph.org/~xiphmont/demo/neil-young.html

TL;DR: encode above transparency level when using the lossy codec, and it won't have audible difference with lossless playback. But again, that's only for playback. As soon as you'd want to re-encode anything, there is no substitute for the lossless original.

I can't for the live of me think of a phrase that gives me Google search results even remotely related to the topic, but I am pretty sure I once found a study that said that consciously inaudible differences in audio due to compression still made a difference - subjects who got the compressed sounds tired more quickly (of hearing them). Having taken a bit of neuroscience I can easily accept that - but I did not take enough (ns) to say for sure that it is so. I can easily accept that merely asking people about anything is not actually objective, it needs more credulity to believe my claim that even people in a blind test who can't tell the difference between two songs (lossless vs. compressed), which is more objective than just asking about qualitative measures, is not a reliable way to determine the question of "lossless vs. compressed".

So could anyone confirm or deny such a study and the mechanism exists from a basis of actual knowledge?

That falls into some speculation area, unless actually substantiated with serious studies. I'd be interested in it, if you can find a source. There are too many hoaxes going regarding audio, to be skeptical enough.

Here is one read on this: http://www.skeptic.com/eskeptic/10-01-06/

    > That falls into some speculation area, unless actually substantiated with serious studies.

I made that quite clear, and I asked for the latter in the prominently placed last sentence. It's the main point of my post, as I think I made clear. I really don't know what more I could/should have done apart from dropping the question entirely, which I don't think is fair - or useful?

You said you saw some study, but can't find it now. I guess if it's a known issue, it will come up eventually.

This sounds familiar. It brings a few things to mind.

I can't remember the term that was used, but I remember reading about audible frequencies possibly being affected by inaudible frequencies in ways that are perceptible to humans. So if you have two source files played using the same equipment with one including the inaudible frequencies then they will sound subtly different due to the interaction.

With the 'getting tired' while listening thing, I know just what you mean. Listening to music on a laptop for instance is a mentally draining experience for me. I've heard it explained like this: your brain knows what a piano sounds like and when it hears the poor imitation it is busy 'filling in the blanks'. Listening on good equipment is much more relaxing - as is listening to something that isn't 128kbps.

I don't claim to know that either of the above are true, but to me they are plausible.

I'm also a little sceptical of the 'lossless is no better than good lossy' claims that inevitably come up in these discussions. While I accept that there is likely a point that audible differences between lossy and lossless would be imperceptible, I've been hearing those claims for a long time (starting with "it's digital - it's cd quality'). When I was seriously interested in these things there most definitely was a noticeable difference. I was right about 128kbps. I'm confidant I was right about 256kbps. Maybe with 320kbps that's changed now, but I don't know as I haven't had a decent stereo for some time. I'm not about to be convinced by those blind test studies that people keep pointing to as objective and conclusive - they always make me think of that one where it was 'proved' that cheap wine is just as 'good' as expensive wine.

I'll stick to lossless when possible and compress to lossy when necessary. Can't go wrong like that!

Yes, I agree. I just think I better point out - for other readers - that my point was made for when people are unable to hear a difference even in blind tests. It was about a longer-term effect that is not part of the direct listening experience. Even if my memory was correct this effect would not change people's difficulty (or at some point, inability) in telling different sound sources apart.

That article is almost entirely about how it's useless to encode audio at excessive sample rates and bit rates beyond the human ear's capacities—not about lossy compression. It does claim that modern lossy compressors are good enough, but admits that there can be reason to prefer lossless distribution since there is then no need to trust the distributor to use a good encoder with correct settings.

Yes. The point is not about that distributor can fail to use a good encoder (anything can happen), but that they can use a good one if they want to, and audible result will be the same as lossless.

When I buy music, I always try to buy it in lossless FLACs anyway. And then I encode it to Opus for playback. But for listening to something on-line, Opus would do just fine to begin with.

Reply on news.ycombinator.com