Interesting Stuff I Found

Thanks to Michael for posting such an interesting and thought provoking article. One thing I have noticed where I have both high resolution and CD versions not on the same disk that the high resolution version is obviously a different master. When releasing high resolution versions of older recordings, particularly with analog masters much greater care was taken in the mastering. It makes the actual comparison between the two versions impossible

The advice to down-res a higher sample rate / bit rate recording to a lower rate to compare sampling rates makes good sense. The hybrid SACD's I have sound very similar on the CD and SACD layers since they came from the same master, however it's impossible to reliably judge with A/B comparisons because of the switching time lag and level differences.

And finally the biggest problem I see with high resolution downloads is the lack of provenance in the recording itself, it's hard to be sure if it's an original high resolution master or an upsampled CD quality recording.

The original recording is the limiting factor in sound quality. I have many CD's with far better sound quality than some of my high resolution recordings simply because the original masters were of much better quality. If the original recording is heavily limited-bad or compressed-even worse, sampling at infinity bit-rates and bit-depths won't improve the sound.

I do think recording at higher bit-rates and sample-rates is beneficial since it's easier to build a good analog anti-aliasing filter with a gradual slope to 48kHz for a 96kHz sampling rate or a 96kHz cutoff for 192kHz than a brick wall 22.05kHz filter for 44.1kHz. The article glossed over the importance of the analog anti-aliasing filter, I don't believe there is any way to fix aliasing products after a signal has been digitized. Early D/A converters also needed a second analog filter on the output to filter high frequency quantization noise (from the stair-step), however, most if not all modern D/A converters use over-sampling to move that noise way up in frequency where a gradual slope analog filter will suffice. As stated a bit depth of 24 or 32 helps prevent digital clipping which is really bad and prevents data loss during mixing.

After all this yapping, I'll say that I do enjoy most of my high resolution recordings. The extra care taken in mastering usually does have a huge effect on sound quality.

Garbage in, Garbage out! I have used the SoX resampler for many years and feel it does the best job of any that are available. Highly recommended.

Without getting into a deep technical discussion, I'll just add that my experience is fairly consistent with what folks are saying...high resolution delivery format alone does not guarantee great sound...and can in fact sound worse. I'm currently listening to a Jazz at the Pawnshop download from HDtracks:



It sounds okay, but it took a long time to load over the network and there were some audible dropouts during the first 15 seconds of playback that I never experience with this setup when playing redbook CD material. I also happen to own versions of this album at 88.2kHz and redbook CD. There are significant mastering differences between the 88.2kHz version (which I believe came from the SACD) and the other two, but generally, the redbook CD version is the most enjoyable to listen to.  :-P

Perhaps the best way to settle this for yourself is to go ahead and buy a well mastered album at 192kHz and then use high quality sample rate conversion software like SoX to downsample the that album to 96kHz, 48kHz, and perhaps a 48kHz, 16-bit version as well. Add a few tracks from each version to a playlist, shuffle, and just listen. Make notes about which tracks had your toe tapping and which left you feeling flat. Should be an interesting experiment.

I am not willing to get into a deep technical discussion in this forum, however you are right in the sense that frequency spectrum being recorded is limited to the instruments being recorded as well as their overtone series and the sounds generated by the room. Where I think his argument falls down, and he doesn't address this, is in the recording process itself. When making a digital recording, the sampling rate determines how many times per second a sample is taken and encoded into a digital "word". The more "words. the more data saved. The range of frequencies in the recording is not the question. The rate of sampling is!

My understanding, gained from a Bell Labs engineer in a class may years ago is that the more samples per time period, the more accurate the reconstruction of the recorded sound will be. Each sample is encoded into a data "word" that then must be decoded. It is decoded (and I am oversimplifying here) by reading the word, in the time box that it was placed, looking at the next word, and then using an algorithm (basically a calculus formula that looks at the two points and determines an arc that will connect the two as music) that fills in the continuum between the two points. My understanding is that the more data points, the easier it is for the DAC to reconstruct the wave and the more accurate that reconstruction will be. Based on my understanding of this process the higher the sampling rate, the easier and more accurate the decoding. If, however, we take a 44.1 signal and "upsample" it, there is no more data than in the original recording. What we have accomplished is to possibly make the decoding more accurate and simplify the job of the DAC. There are other associated issues to discuss such as low volume level recorded information, digital noise, and phase anomalies caused by jitter.

I believe that I can hear differences when I upsample 16/44.1 files to 24/88 or 24/96. I believe that what I am doing makes it easier for the DAC to decode the signal and reduces the phase and timing anomalies produced by the decoder. (This is even more impactful if a highly accurate clock is in use in the decoder) However, unless a source was recorded at 24/192 and then provided to the purchaser at that same format, the possibility of a wider frequency response is not possible... if it wasn't recorded, it can't be generated later! I do agree that 24/192 or even 24/176 are not realistic for storage and reproduction of music files. As described in the article, there just isn't enough music at that high a frequency or dynamic range for it to be worth the storage size of the file.

This is strictly my personal opinion and does not reflect the official position of the Audio-Video Club of Atlanta.

John

I skimmed his article, and something doesn't make sense to me. Doesn't the sample rate just refer to the number of times the codec samples the signal? I don't see why that translates to extending the frequency response of the output? I may be way off here, as I didn't read his whole article.

Ya but he doesn't quantify the effect of having Hospital Grade AC power plugs !!! ;-)

And little bridges to hold the speaker wires off the nasty floor ;-)

OK, seriously, has anyone objectively verified superior sound from 24/192 vs 16/44 in blind testing? Neil Young says there is a quantifiable difference, I love him, but not sure I believe his belief on this. For sure he is correct in saying it will be much better than most commercial compression methods, MP3, AAC, etc.

I ran across this article today and wanted to share this with the group. This article came at a great time as I recently started a wishlist on HDTracks and was looking for the 24/192 versions of anything that I was interested in.

In short, it states why 24/192 sounds worse than 16/44 and that it consumes 6x the amount of space. Although not mentioned, these higher res downloads cost more.

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

Michael Dean