Hacker News new | ask | show | jobs
by nominated1 1941 days ago
Purely marketing for the “more is better” pseudo audiophile.

ABX testing blew the “lossless sounds better” mantra out of the window nearly 2 decade ago [1]

[1] https://en.wikipedia.org/wiki/Codec_listening_test

7 comments

Doesn't matter how high quality spotify is if new music is mastered so poorly.

Also doesn't matter for me since my car, a tech filled tesla, streams spotify at what sounds like the lowest possible bitrate.

Yes. This is a major problem. Shit in shit out. I'd love to revist a lot of songs if someone would be able and willing to master them skillfully. Low dynamic range is the main culprit here.

Too low bite rate can show in some few cases. High hats seem to be difficult and some organ music, transients in general iirc. And you need good enough equipment (I don't mean expensive). This kind of analytical listening to detect small problems is exhausting. Relaxed listening does not need lossless codecs.

But mastering is the key. I would pay premium for good mastering instead of lossless.

>Cunningham and McGregor [1]

>A total of 100 participants engaged with the listening test and were recruited from the Merchiston campus at Edinburgh Napier University. With respect to background, 28% were students at the University, whilst 33% were academic or faculty staff and 39% were administrative and support staff. Participants were not offered any form of remuneration or any other form of inducement for their involvement.

This [1] Was the only test that had included Uncompressed WAV or PCM as anchor. And it was tested with sort of randomly selected participants, and randomly selected sample sound track.

>Purely marketing for the “more is better” pseudo audiophile.

All the other test listed were done during what I called the Audio Codec era on HydrogenAudio. Where professional encoder developers and enthusiast ABX the hell out of psychoacoustics tuning. Along with problematic samples testing.

For 90%+ of general public, MP3 128kbps with a decent encoder ( LAME ) has been good enough for well over a decade. Suggesting better codec doesn't matter and is purely for pseudo audiophile completely neglect the work people have been tuning and making these audio codec better for the past ~20 years.

And yes, 10 years later Musepack v8 still beats AAC or Opus in High Bitrate.

[1] https://www.hindawi.com/journals/ijdmb/2019/8265301/

> Suggesting better codec doesn't matter and is purely for pseudo audiophile completely neglect the work people have been tuning and making these audio codec better for the past ~20 years.

I never suggested that better codecs don’t matter. I was stating that lossy codecs are amazing and that lossless, no matter which codec is used, isn’t necessary.

It is more interesting in that it shows that Spotify isn't worried about people ripping audio that is in theory more valuable for archival.

(I can't argue based on perceptual differences, but it just seems "wrong" to have the compressed version be the library master.)

You make a claim, but provide a link to a page that does not support your claim. Specifically, the "public results" section shows the winner of comparisons between compressed codecs, and not whether those codecs achieved transparency.
Correction: there is one footnote on that page, and it links to a single study performed by a team creating a compression algorithm that found that compression algorithms sound just as good.
My only question was why did they wait so long? They could have rolled out 321kbps super-duper mode for $14.99 years ago. People would have paid for it.
I can guarantee it’s not. A simple comparison between Spotify and Apple Music shows you with no doubt that Apple is doing something better with regards to compression, it sounds a lot more defined, so there is clearly room for improvement for Spotify.
> … shows you with no doubt that Apple is doing something better with regards to compression…

No it doesn’t. It shows their mastering/tuning techniques are different or they have access to different sources. Compression, wrt codecs, has nothing to do with the difference in sound you’re hearing.

Send me to hell if Spotify is mastering or applying EQ to any tracks. That would be egregious.

It’s definitely about compression. Implementations can differ significantly for the same formats, as anyone who remembers the initial MP3 frenzy can tell. The fact you cannot discern two random codecs at their max quality doesn’t mean Apple and Spotify cannot have a meaningful difference with their streaming codecs (both custom afaik).

> Send me to hell if Spotify is mastering…

Pack your bags. Just Google it, or you can start here [1]. A relevant quote “This is because Spotify applies Loudness Normalization to your tracks as they're played to listeners”

A brief search shows that both Apple and Spotify offer 256kbps AAC. ABX testing these two would be pointless. They’re both above transparent. It’s NOT the codec.

Send me to heaven if X would allow for parametric EQ settings based on headphone/sound system. This would appeal to *actual audiophiles [2]. This would confuse the hell out of normals, so bullet point marketing is what we get.

[1] https://artists.spotify.com/faq/mastering-and-loudness [2] https://github.com/jaakkopasanen/AutoEq

It sounds to me as though you're saying that Spotify's loudness normalization is affecting the audio quality of the records. I don't believe this works the way that you may think it is working.

While Spotify IS applying file compression (ala mp3) to every one of their records (and this affects the sound quality) I have not heard any reports of them applying any additional post-EQ or volume compression in the way that, say, a radio station might. I believe you may be mistaken as to what the Loudness Normalization feature is doing.

It is simply turning down the volume of tracks that have been mixed/mastered to excessive loudness levels so that they match the average loudness of more sensibly-mixed/mastered songs. Turning the volume of the louder track down has no effect whatsoever on it's audio quality. The meta-effect, then, is that the normalization disincentivizes the practice of mixing/mastering songs too hot; something which DOES have a negative effect on the audio quality of a song.

Loudness normalization happens on the client, and can be opted-out.

What is it then? At any rate, if this new offering can match Apple Music quality then it’s a win - and not “pure marketing”.

There is a cut and dry relationship between bitrate and audio fidelity. You simply cannot encode higher frequency audio in lower bitrate samples.

https://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampli...

That's irrelevant because human ears aren't spectrum analyzers. Audio compression codecs exploit weakness in human perception[1] to discard data with minimal loss of subjective audio quality.

[1] https://en.wikipedia.org/wiki/Psychoacoustics

Too bad we have that pesky brain in the way. Human perception isn't just a matter of signal processing theory.
Nyquist is about sample rate and frequency, not bitrate and frequency.

Edit: Clearly there is generally a relationship between bitrate and quality, but for compressed audio it is far from "cut and dry".

Bitrate is just samplerate * bit depth. For 44100 and 32 bit float you have 1411200 bit per second.
That's only true for uncompressed audio.
Yes, but for most codecs, bitrate is variable. For the parts where the higher frequencies are present the codec is free to bump up the bitrate and it can also scale it down for silent parts or parts with low frequencies only.
Sure, if it were encoded at a variable rate. But then it wouldn’t be 320kbps CBR. Normally when I see people refer to 320kbps audio they literally mean constant bitrate. If it’s variable then for LAME mp3 people would specify V0 or V2. At least that’s the taxonomy that I absorbed when I was active on what.cd

Edit: you have edited your comment to remove mention of 320kbps so my comment is now moot :)

Yes, but afaik Spotify uses Ogg Vorbis, which is inherently VBR. https://en.wikipedia.org/wiki/Vorbis
Sorry. I removed the part about 320 kbps being an average, since I realized it was wrong, but you were faster :D.