• Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Talking to AI about Lossy Codecs

shrunk_thyroid

New Member
Joined
May 7, 2023
Messages
4
Likes
1
Hi, new user here,
I'm no expert, just someone who reads a lot and tries to understand why stuff sounds different to his ears. A few days ago, I had the probably stupid idea to ask Claude AI about one of my pet issues - why certain music sounds so different on Spotify (with highest quality settings and normalization set to low), compared to Qobuz. My normal understanding ist that Spotify should sound as good as Qobuz, as ogg vorbis should be transparent. Sadly, my ears disagree, even listening to bluetooth headphones using the LDAC codec.
This is what Claude AI came up with:

User: I think classical piano on Spotify sounds terrible, as if the strings were muted. On YouTube Music it's better, even though they also work with lossy codecs. Am I imagining this? Same effect with classical guitar, among others.
Assistant: You're not imagining it - there's actually something to that! Several factors can explain this:
Different audio codecs and bitrates: Even though both services use lossy compression, they often use different codecs (AAC, Ogg Vorbis, etc.) and different bitrates. YouTube Music sometimes uses higher bitrates or other encoder settings that work better with classical instruments.
Mastering and source material: The platforms often get their music from different sources or use different master versions. Some labels deliver specially optimized versions for certain platforms.
Audio processing: Spotify applies additional volume adjustments and normalization that can be problematic for classical music. These automatic processes can compress dynamics and make the sound "flatter."
Particularly vulnerable are instruments with complex overtones like piano and classical guitar. The fine harmonic details and natural decay of strings suffer greatly under aggressive compression.
You could try setting Spotify's audio quality to "Very high" and disabling "Volume leveling" in the settings - that sometimes helps. But basically you have a good point: for classical music, it's often worth comparing different streaming services.
User: Everything is already set to highest quality. I thought ogg vorbis from about 200 KB/sec is transparent.
Assistant: You're right - Ogg Vorbis from ~192-256 kbps is normally considered transparent. But with classical music it's more complicated:
Classical music is particularly demanding: The enormous dynamic ranges, complex harmonies, and natural decay of piano strings challenge codecs extremely. What sounds transparent with pop/rock can still have audible artifacts with a Steinway grand piano.
Encoder tuning: Not all Ogg Vorbis encoders are the same. Some are optimized for rock/pop, others consider classical music better. YouTube could simply use better encoder settings or newer versions.
Pre-echo and ringing: With percussive instruments like piano, codecs can create "pre-echo" - a slight "ringing" before the actual attack. This can sound like muted strings.
Psychoacoustic models: These are often calibrated for "normal" music. With classical instruments with their long reverb tails and subtle overtones, they can make wrong decisions about what's "inaudible."
It's quite possible that YouTube Music simply uses more conservative compression settings or better encoders for such critical content. Your hearing isn't deceiving you - with classical music, the differences between platforms are often more apparent than the bitrate alone would suggest.
So is this another example of crazy AI, or is classical music really too hard for some lossy encoders? I thought that electronic music was actually harder to digest for them.
 
Are you using Spotify Premium? On what platform? It looks to me like Spotify uses rather low bit-rates for all non-Premium members, and even then only the Very High setting gets you to 320kbps which is generally the threshold where it should be generally impossible to tell the difference. The 256kbps on the web browser should also be OK. But 160kbps or lower? Definitely not impossible that artifacts might pop up.

I also don't see any mention of Ogg Vorbis. Seems they're using AAC, at least for the web player. Should be fine either way.

1749052045678.png
 
Occam's razor postulates that an obvious explanation is probably the correct one. In your case, I would not bother with codecs transparency issues before I exclude more obvious factors such as different masters and different normalization settings (Qobuz does not use any normalization, AFAIR, and even if it did, there are little chance they would use the same processing as Spotify does).
 
Yes, I use Spotify Premium, and I made sure that I have the quality set to very high quality. They use AAC for the web player, but OGG Vorbis for the app, up to 320 kbit/sec. I also made sure that Spotify cannot downregulate my music. Still, especially classical piano and guitar sounds cheap to my ears, compared to Qobuz.
For a popular example, even though it's not classical music, you can hear it in Keith Jarrett's Köln Concert. But as not more people are complaining, it could be me and my wild imagination. I'm open to that possibility.
 
Occam's razor postulates that an obvious explanation is probably the correct one. In your case, I would not bother with codecs transparency issues before I exclude more obvious factors such as different masters and different normalization settings (Qobuz does not use any normalization, AFAIR, and even if it did, there are little chance they would use the same processing as Spotify does).
My understanding of normalisation on Spotify is that up to medium, it should not interfere with sound quality and it should not "compress" the sound, but I'll have a go at it..
 
it should not "compress" the sound
How does it change the loudness then?

And if we talk about volumes, did you make sure you were listening at matching volume levels? Going further, did you exclude any bias by conducting blind tests? All these factors can influence the sound perception not any less than lossy compression.
 
QObuz ist a lot louder than Spotify for many tracks, and I adjusted volumes manually. It's quite possible that this ist where my bias comes from.
It's would actually be a relief for my wife If we could finally stick with Spotify.
 
I think the first response from the AI seems reasonable, but I think the second bit mostly about lossy codecs sounds suspicious. Do we really think the creators of OGG forgot classical music and pianos exist?

If there really is a sound quality difference in the same content between the two platforms, (not masters etc) I'd suspect mishandling on Spotify's part (transcoding from a lossy copy at some point, who knows) over an actual deficiency in the codec.
 
Back
Top Bottom