Makestation

Full Version: Spotify "Data Saving" mode
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
I discovered this recently, and it's been interesting. I don't have unlimited data, so saving data on Spotify is a big deal for me. I didn't realize this, however: It's 24kbps (HE-AAC). Spotify normally uses Ogg Vorbis, so I'm surprised they dropped it to AAC for this, but the 24kbps is shocking, to say the least. 

I took a listen, and it's absolutely amazing how good it sounds at such an abysmal bitrate. Make no mistake, I can definitely tell the difference from the higher quality ones (it sounds a little more reverb/echo-ish, especially when the drums come in), but it doesn't sound terrible by any means. 

Sirius/SiriusXM/XMRadio receivers generally use VBR at around 30-44kbps and sound MUCH worse (especially the old Sirius-only radios, the newer XM and SiriusXM ones sound a little better), so I'm quite surprised. 

I generally can hear the difference between an MP3 and the CD at anything about 160kbps and lower, or with AAC, I can hear the differences even between 256kbps and the CD (surprisingly, as AAC is a better format).I'm not even an audiophile (haha Finna), and can definitely hear artifacts in higher bitrates, so I was surprised at how good Spotify pulled it off at 24kbps. 

Anyone else used low bit rate audio before? What were your experiences? It's more or less obsolete with the storage we have available now, but it's still amazing how well some of the more modern formats can pull this off.
I use Pandora, and I'm not aware of a specific data-saving mode, but I just use the offline mode so I don't use it at all. Real nice when working in areas of low signal.
Ah, nice @Guardian. Yea, offline is the way it always used to be done. I sort of miss it. I hate not being able to play music without a data signal. Finna

---------

I did a pretty in-depth listening test across a variety of formats and bitrates. The results were surprising: Ogg Vorbis is an absolutely fantastic format. It outperformed MP3 and AAC at every single bitrate I've tried it at. And profoundly (and very noticeably) so.

The contrast was so extreme that I'm surprised Vorbis has remained an obscure format. Spotify uses Ogg for most of their streaming and some video games use it. Besides that, it's rarely supported on most platforms, and it's really a shame. Ogg was, hands down, the best format I tested (particularly at lower bitrates) by a landslide.

I could even tell the difference between 192 kbps MP3 and the Wav (although it was extremely subtle, I would have never been able to tell in a blind test without comparing them back and forth constantly. It was particularly the crash cymbals and some of the tom-toms on the drums that were most noticeable). Ogg Vorbis at 192 kbps and 320kbps were both completely indistinguishable from the CD quality audio in every single way.

Surprisingly, AAC was worse even at 256 kbps. It sounded very good in terms of fidelity and general quality, but there were more audible artifacts across the crash/cymbals than there were even on MP3. Perhaps my ears are trained to hear it, but AAC has always had audible artifacts to my ears, even at higher bitrates. Finna

In any case, Ogg Vorbis started to sound okay at about 40kbps (amazingly) and went up from there. At 96 kbps, it sounded almost as good as the CD with a few artifacts, and at 192 kbps, it was completely indistinguishable. Neither AAC nor MP3 sound even listenable below 128kbps, so the fact that Ogg Vorbis sounded passable at 40kbps and decent at 96kbps was quite impressive.

That being said, Ogg doesn't explain how Spotify pulled off 24kbps on their data saving mode. Not even close (and it's so low that even phone calls sound bad at this bit rate). Ogg, as a matter of fact, sounded pretty bad at 32kbps (despite sounding okay at 40kbps). I did some more research, and I discovered why Spotify chose AAC-HE for this one. This codec is designed specifically for this very purpose: For extremely (and laughably) low bitrates. In fact, it is designed and tailored so well for that purpose that it actually sounds WORSE at higher bitrates. That's literally just how well this format is optimized for low bitrates.

As it turns out, AAC-HE achieves it by not actually encoding high frequency information at all. It pretty much just chops off anything above 10khz, and uses SBR (spectral band replication) to "guess" on the frequencies it removed by looking at harmonics (the same frequencies at lower octaves or that resonate at lower pitches) to figure out what frequencies need to be added back in.  In other words, it just focuses on lower frequency/pitched information, encodes it as well as it can in the 24kbps, and reconstructs everything else that it flat-out refused to encode during the decoding stage. It's quite genius, really.

More information:
[spoiler]
Technically, this is a huge oversimplification. SBR cuts the entire sample rate in half, resulting in the physical inability to encode any pitch/frequency above 10-12khz. However, this gives significantly more headroom to accurately encode the remaining audio data. SBR then uses harmonics from the lower frequencies to reconstruct higher frequencies, but this process results in more inherent artifacts and errors. As such, AAH-HE uses a separate data stream with additional "control bits" to provide "corrections" to this guessing model to compensate. It is essentially guiding the decoder to guess appropriately during the encoding process, and it specifically chops off additional frequencies that are most easily guessed/reconstructed. AAC-HE is perhaps one of the most advanced audio formats in existence because of this. Very few formats can reliably encode music at such laughably low bit rates. 

Even still, it has its disadvantages. AAC-He is rarely used for higher bitrates because of the artifacts that result from guessing the higher frequency information. At 128+kbps, it makes much more sense to simply encode these frequencies normally, and doing so results in a less artificial reconstruction of the full audio spectrum. 

More information: http://www.stockis.se/AES112th_SBR.pdf
[/spoiler]


It's incredible what they pulled off in 24kbps at Spotify. Technically it's cheating (SBR simply flat out guesses a lot of information), but it sounds good (for what it is... Finna). At higher bitrates, Ogg Vorbis rules them all. By a landslide.