Actually, the discussion started with that topic. IMHO issues like resampling, dithering and declipping go hand-in-hand with zero-padding.
Zero-padding alone will not influence soundquality too much. Timing stays the same and the digital steps stay the same. So only differences in handling the 16bit and the 20/24/64float coding by the DAC will influence the soundquality.
More interesting to me is the ability to dither, resample and declip (sorry, I happen to like music that is usually recorded the wrong way, but I do not like the digital clipping). Yesterday, searching for software to declip audio, I came upon iZotope RX3 and installed the demo. I fed the program with a beautiful song (clipped) and played around.. after dithering, resampling and declipping (in that order) I got a very good result. The harshness of the clipping had gone, more detail and better soundstage in general.
The problem was $349,- to be able to save this to disc...
IEEE754 64 bit floating point encodes without loss integers up to about +/- 2^53.
With the usual 24 bit audio sample range, the conversion to double (or float) and back to integer is exact.
With 32 bit integer samples, only double precision would be lossless.
This is also true if the range -1,1 is used for the float encoding. Of course any floating point operation like gain will introduce rounding errors...
But, 32 bit samples are easier for 32 bit processors than 24 bit samples (which impose some byte shuffling), so YMMV I guess...
Storage in a higher coding than 24bits is only justified if you plan future digital processing or reproduction on a 32bits DAC. Otherwise the extra bits only are a waste of space.
If I understand you correctly this is only beneficial if the DAC would upsample
per default, correct ?
With a non upsampling DAC (but capable of higher sampling rates of course) would
it still make sense (give a cleaner output spectrum) to feed it with data that has been
upsampled by software ?
Julf, it´s not about "what matters to me", but about measuring an effect.
I think my question (which was addressed at Pavel who has obviously
extensive experience in measuring such things) was rather clearly stated.
There is no need for you to interpret what "matters to me".
Thats why I mentioned ARTA. Sometimes it helps to actually read the posts before responding...
Thats why I mentioned the ADI-2 I use (which is a bit better than the usual sound cards, BTW).
Sometimes it helps ...
Georg,
please check this post
http://www.diyaudio.com/forums/pc-based/251845-over-sampling-44-1khz-wav-files-2.html#post3835811
I have placed it in the 44.1/16 oversampling thread, as it is directly related to the subject there.
Regards,
Pavel
please check this post
http://www.diyaudio.com/forums/pc-based/251845-over-sampling-44-1khz-wav-files-2.html#post3835811
I have placed it in the 44.1/16 oversampling thread, as it is directly related to the subject there.
Regards,
Pavel
We can measure a zillion different things. Some of them reflect subjectively perceived sound quality, but most of them don't, as long as they are within reasonable limits. Thus it is a question of what matters to you if you choose to care about a difference that might or might not be audible.
Our measurement instruments are great for measuring objective differences, and verifying our designs and models, but in the end it is the subjective sound quality perceived by the listener that is the ultimate "measurement".
iZotope
There are softwares using iZotope technology at much lower price. Please check this link. Their Sample manager and Triumph use iZotope technology.
Audiofile
Yesterday, searching for software to declip audio, I came upon iZotope RX3 and installed the demo. I fed the program with a beautiful song (clipped) and played around.. after dithering, resampling and declipping (in that order) I got a very good result. The harshness of the clipping had gone, more detail and better soundstage in general.
The problem was $349,- to be able to save this to disc...
There are softwares using iZotope technology at much lower price. Please check this link. Their Sample manager and Triumph use iZotope technology.
Audiofile
Minimum phase filtering using SOX
Okay, I found out how to do the kind of processing I was looking for.
So starting with a 16-bit Redbook CD, I'm using foobar with the SOX plug-in to do the up sampling. I have been up sampling to 88 kHz and 176 kHz at 24 bits FLAC files. In particular, I wanted to compare normal up sampling to minimum phase filters.
The 176 kHz files are huge, and I presume they should be saved for special circumstances. They sound good, but to start with I mostly listened to 88 kHz files. In both cases up sampling in advance seems to be an improvement.
However, the minimum phase upsampling seems better, and seems more musical, and with fewer "digital" artifacts. Of course, we are not achieving SACD quality, but more of the magic seems to be there when using the minimum phase filters.
In particular I set SOX to bandwidth of 90% and minimum phase, with anti-aliasing, and best quality. Apparently, there is an updated plug-in that will allow you to go to 87.5% bandwidth which may be even better.
Eric
Okay, I found out how to do the kind of processing I was looking for.
So starting with a 16-bit Redbook CD, I'm using foobar with the SOX plug-in to do the up sampling. I have been up sampling to 88 kHz and 176 kHz at 24 bits FLAC files. In particular, I wanted to compare normal up sampling to minimum phase filters.
The 176 kHz files are huge, and I presume they should be saved for special circumstances. They sound good, but to start with I mostly listened to 88 kHz files. In both cases up sampling in advance seems to be an improvement.
However, the minimum phase upsampling seems better, and seems more musical, and with fewer "digital" artifacts. Of course, we are not achieving SACD quality, but more of the magic seems to be there when using the minimum phase filters.
In particular I set SOX to bandwidth of 90% and minimum phase, with anti-aliasing, and best quality. Apparently, there is an updated plug-in that will allow you to go to 87.5% bandwidth which may be even better.
Eric
No, I didn't try that, I don't see any reason to restrict ourselves to 16 bits. The methods I'm discussing here obviously require increasing bit depth.
There are plots of the impulse response of these filters here:
Transporter Round Two [SOX] (was Transporter for £999...) - The Hitchhikers Guide To Meridian
In the second half of the page.
These techniques are being used by Meridian and Ayre in their players.
http://www.ayre.com/white_papers/Ayre_MP_White_Paper.pdf
There are plots of the impulse response of these filters here:
Transporter Round Two [SOX] (was Transporter for £999...) - The Hitchhikers Guide To Meridian
In the second half of the page.
These techniques are being used by Meridian and Ayre in their players.
http://www.ayre.com/white_papers/Ayre_MP_White_Paper.pdf
Last edited:
With sox you can't mimic the impulse response of either Meridian nor Ayre. Only some minimum and immediate phase things. I may suggest creating spectral images of these recommended impulse resonses and wonder how much these change the whole frequency behaviour. Better to have the standard pre-echo at frequencies you can't hear.
Edit: i once did some for downsampling with sox but it should illustrate what i mean
Attachments
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.