[gkefalas]

Hey all, I did a technical breakdown of high-fidelity audio recently for our company and thought that it might be of interest to hackers. I'm especially interested in ideas & criticisms to lead to a more thorough test methodology for when we revisit this topic in the future. I know I simplified or explained away a lot of complex subjects in digital audio, but otherwise this would have been an even more massive post!

[com2kid]

It is a good analysis! The null test was surprising, I'm sort of interested how bad the result would be if you went up against something like 192kbit MP3! My music collection is in FLAC and MP3 (256kbit VBR actually), just because MP3 is so widely supported.

Thank you for the interesting read!

[byproxy]

In my own testing the only thing that stands out in null testing is artifact noise and stuff that's lost in the higher frequencies, i.e. cymbals. The more compressed you go, the more artifact noise introduced and high frequencies lost. As someone who can enjoy classical music (for simplicities sake "classical" and not some other classification), I notice that a bunch of "audiophiles" seem to think that's where their money is going - to increased listening satisfaction of classical - when that tends to be the type of music where least is lost in compression. Music with constant cymbal hits (e.g. rock music with drumkit) seems to lose the most.

Again, my own personal null testing with music spanning various genres in my library.

[gkefalas]

Thanks! I agree, some things were surprising. I'm really bummed that the null tests were affected by source audio amplitude, plus the frame size causing alignment issues—I feel like it's not as accurate a test as it could have been. I have to revisit the methodology to come up with a baseline of sorts. I was thinking of doing a comparison all the way down to 128kbps next time… sounds like what you're saying!

[davidgerard]

Audio woo is always worth debunking, this is pretty good.

I wonder if Tidal will try to pivot to selling pure snake-oil 24/96.

[gkefalas]

The amazing thing is that they probably could have made a go at it focusing on paying out more to artists (despite the fact that it's not the fault of competing services for negotiating the best rates they could with the labels). Or, maybe making a go at the sound quality argument but priced the same as Spotify Premium. Like you, I wouldn't be surprised if they eventually go 24b/96K, but that will be really difficult since that would drastically limit their catalog.

[davidgerard]

Yeah, let's convince Nicki Minaj fans they need 24/96 ...