Testing and feedback of new “Hi-Res” digital remastering algorithm

Hi, I am new to this forum, and I would like to seek for testers and feedback of the digital remaster algorithm I built.

<<Introduction>>

While the term “hi-res audio” (that golden logo) has largely just marketing buzz to sell more expensive digital music, I also met many people 

in person that clearly appreciate the differences. Personally I think the mixed results originated from inconsistent remaster process. 

The conventional way of remastering from tapes with new digital exports increase the cost for record companies, who then pass it to the consumers. 

For the past few years I have been working on how to programmatically remaster (not just upsample) CD quality WAV into “hi-res” format matching JAS (Japan audio society) description. While started as skeptics, my curiosity was largely provoked by the article “The world beyond 20kHz” by David E Blackmer. It maybe also due to my previous background working on tomographic algorithms, which tries to reconstruct complete information out of limited measurements — similar to this algorithm I built for audio.

<<Comparison>>

Today, anyone can use software to easily upsample a CD WAV into higher resolution, but the problem is: besides changed format that supports higher sample rates, nothing has really gained, as shown in the following two spectrograms: on picture (1) shows the original CD WAV file (44.1 kHz / 16 bits), and picture (2) is the graph from SoX rate conversion into higher resolutions (192 kHz / 24 bits)

picture (1) https://1drv.ms/u/s!Aqkw0or3U7WtbGphQHoDrmfpSRc

picture (2) https://1drv.ms/u/s!Aqkw0or3U7WtamphQHoDrmfpSRc

As far as I know, many music tracks sold online today claiming “HD” are actually just going through the same process. Here we can clearly see none of the extra frequency provided by higher resolution format was actually utilized. In fact if we changed the scale of the graph from CD (pic1), we can get the same result as the upsample one (pic2). 

So, what would my algorithm do to the same song that I just showed? Please take a look at the following spectrogram:

picture (3) https://1drv.ms/u/s!Aqkw0or3U7Wta2phQHoDrmfpSRc

A few things to point out: 

1. The image scale is the same as the sox upsample graph I showed previously and the resolution is also 192kHz/24bits.
2. The algorithm reconstructed frequencies up to around 48kHz, which covers the 40kHz range suggested in both David’s article as well as JAS definition of “Hi-Res”
3. High pass triangular dither is used in final down-mix.

<<Sound>>

So, how does it sound?

I have been using the same algorithm for the pass three years and converted hundreds of CDs from different genres, and I am please to say that when paired with capable equipments, the results are unprecedentedly satisfying: not just because I enjoyed them, but the guests and family visited my place were sometimes shocked how much better that familiar songs now sounds better! 

Due to copyright concerns, I cannot just share the file publicly. But I am seeking people that are interested to test this algorithm. If you are interested, please email me at:

[email protected]

<<Caveats>>

Like any algorithms, of course, this approach has its own drawbacks:

1. It is slow and resource hungry: As expected with reconstructions from finite data, it takes a long time, a typical song usually takes about 4-5 hours to complete.
2. It is, in the end of the day, pure computational audio process that does not involve “remastering” in traditional sense.

One of the reason that I waited 3 years before seeking public feedback is that I believe modern cloud infrastructures has finally ready to process this algorithm with reasonable costs. 

<<Call for testing & feedback>>

As mentioned earlier, I am seeking people that are interested to test this algorithm. If you are interested, please email me at:

[email protected]

Thank you for your attention.