Digital filter "game"

[bennetng]

But was that the case with the Mahler sample where all software clipped except SoX? Unfortunately the source DSD file is not available on the blog any longer.

No, clipping on the Mahler was caused by the implicit PCM gain settings in Saracon and DSD Master. Glitches caused by DSD overloading won't show up as a vertical line straight through the spectrogram because DSD glitches caused by this will be filtered by the PCM low pass filter, you can't easily see it in the spectrogram. If you listen to it or looking at the waveform you can find them much easier.

 

[hyperknot]

I see. So if I manually used no gain setting in those programs that clipping wouldn't have happened. And the price of this is 23 bit of resolution instead of 24 bit right?

 

[bennetng]

Yes. SoX can only report clipping caused by SoX itself. It doesn't know and cannot cure any inherent DSD overloading. If SoX doesn't show warnings, then you are fine.

 

[hyperknot]

What is the reason that you used pipes with your command? Was it because the dsd build was old and sox was an up-to-date version?

I'm trying the new SoX builds from https://audiodigitale.eu/repo/sox/ and I'd like to try to make it a single command, but I cannot make it. I thought that -p is for pipe, but if I remove it then it breaks.

 

[bennetng]

The DSD-capable SoX executable that mansr posted does not support flac encoding and spectrogram generation. If there is an all-in one executable then piping is not needed.

 

[hyperknot]

The DSD-capable SoX executable that mansr posted does not support flac encoding and spectrogram generation. If there is an all-in one executable then piping is not needed.

The non-windows builds from the above link are from 2021 and it seems to support everything. My only problem is that I cannot figure out the command line. Where should "rate" go? In gopts, fopts infile, fopts outfile or effects?

This doesn't work:

soxdsd -V sample.dsf rate -b 74 -v 88200 sample_sox1.flac

 

[bennetng]

@mansr may offer some help.

 

[mansr]

Something like this should work: sox in.dsf -b 24 out.flac rate -v 88200

 

[hyperknot]

@mansr this works but produces an 8-bit file:
soxdsd -V sample.dsf sox1.wav rate -b 74 -v 88200

 

[hyperknot]

Oh, you've already answered. And how do I specify `-f hq` or similar on this?

 

[pLudio]

sox -SV4 in.dsf --bits 24 -C 8 out.flac rate -uL 88.2k gain 6 dither -S

@mansr may have a better suggestion for the dither. -u is the secret ultra quality.

 

[mansr]

There's no need for dither with 24-bit output since there's enough noise already in the source.

 

[hyperknot]

There's no need for dither with 24-bit output since there's enough noise already in the source.

But isn't dither needed when going 32-bit -> 24-bit? How does noise help that?

 

[mansr]

But isn't dither needed when going 32-bit -> 24-bit? How does noise help that?

Dithering at 24 bits makes no difference if the content already has noise at 20 bits, say.

 

[bennetng]

Yes. I have expected comment like this as there are always different camps of people say 24-bit dither is needed or not, just like somewhere else people are talking about is it necessary to use WAVEFORMATEXTENSIBLE or not for a 24/96 file if it is only stereo.

IMO, in the case of generating 24-bit test signal to test something, then I will use dither.

 

[hyperknot]

I see. So the same way trying to determine the right gain at encoding is also silly, as it's just best to leave it without gain and use RG in player, isn't it? So basically no gain, no dither on real world music.

-b 74 is the only question I'm left with.

 

[pLudio]

The FLAC specification allows 20 bit files but it is unfortunately not yet implemented. We can limit the data to 20 bits anyway and save some space.

sox -SV4 in.dsf --bits 24 -C 8 out.flac rate -uL 88.2k gain 6 dither -S -p 20

Saved me 35% on a file.

 

[bennetng]

The FLAC specification allows 20 bit files but it is unfortunately not yet implemented. We can limit the data to 20 bits anyway and save some space.

sox -SV4 in.dsf --bits 24 -C 8 out.flac rate -uL 88.2k gain 6 dither -S -p 20

Saved me 35% on a file.

The same simulation I did in one of the posts on the first page:

Digital filter "game"

Some of you probably have seen a lot of filter plots like these on DAC datasheets: How about making these things without using specialized software like Matlab, Octave and such? They are not particularly easy to use. So many people just "see" them rather than try to figure out the meaning...

www.audiosciencereview.com

 

[bennetng]

I see. So the same way trying to determine the right gain at encoding is also silly, as it's just best to leave it without gain and use RG in player, isn't it? So basically no gain, no dither on real world music.

-b 74 is the only question I'm left with.

-b 74 is the lowest allowed setting on the rate effect. Depends on the target sample rate, it will for example, remove some audible frequency if your target is 44.1k, but for example at 88.2k it won't, and the removal of extra frequency (DSD noise) can reduce file size as well.

Also, if 24-bit dither is useless, than rate -u is useless as well. The added filter length and attenuation are placebo anyway.

 

[mansr]

Yes. I have expected comment like this as there are always different camps of people say 24-bit dither is needed or not, just like somewhere else people are talking about is it necessary to use WAVEFORMATEXTENSIBLE or not for a 24/96 file if it is only stereo.

One of those things is a matter of mathematics, the other about an ill-defined file format. Only the latter is really up for debate.

IMO, in the case of generating 24-bit test signal to test something, then I will use dither.

If generating a synthetic signal with no noise, sure. A DSD source will always have noise well above the 24-bit level.