Our church has been recording music at 96/24, and it sounds very good. We plan on distributing it on DVD at 96/24. But we also need to produce CDs at (ugh) 44.1/16.
All of the resampling algorithms and programs I have tried so far add noticeable (and objectionable) artifacts to the sound ... our music director described it as a "high frequency sizzle". It's least obvious on highly tonal sounds (a solo instrument), and worst on massed choral singing. I'd call it a lack of coherence -- a loss of "palpable presence" -- a bit reminiscent of stylus mistracking on an LP.
The best-sounding resampler I've found so far is Resample 1.1 by SoundsLogical -- polyphase filtering with choices of dither. The worst-sounding one is the "high quality" resampler built into Cakewalk Sonar 3.1 -- sinc method.
Any suggestions?
Thanks,
Rainer
All of the resampling algorithms and programs I have tried so far add noticeable (and objectionable) artifacts to the sound ... our music director described it as a "high frequency sizzle". It's least obvious on highly tonal sounds (a solo instrument), and worst on massed choral singing. I'd call it a lack of coherence -- a loss of "palpable presence" -- a bit reminiscent of stylus mistracking on an LP.
The best-sounding resampler I've found so far is Resample 1.1 by SoundsLogical -- polyphase filtering with choices of dither. The worst-sounding one is the "high quality" resampler built into Cakewalk Sonar 3.1 -- sinc method.
Any suggestions?
Thanks,
Rainer
I haven't tried it for downsampling, but from a mathematical/theoretical perspective I beilve that ssrc available from shibatch.sourceforge.net is pretty close to state-of-the-art. It is a frequency-domain approach and so only certain ratios are supported, but 44.1/96 is certainly possible.
Free and easy to try, so you can't go wrong.
Free and easy to try, so you can't go wrong.
Thank you!
I found SSRC with no trouble, and have tried it on two of my most difficult files. It's better than everything else I've tried so far!
Of course, it's not quite "perfect", so I'm still interested in other suggestions (even those costing money). But I'll start using SSRC, and it will improve my CDs. It's also fast -- about 6X real time speed. (Resample was 0.8X).
Thank you!
Rainer
I found SSRC with no trouble, and have tried it on two of my most difficult files. It's better than everything else I've tried so far!
Of course, it's not quite "perfect", so I'm still interested in other suggestions (even those costing money). But I'll start using SSRC, and it will improve my CDs. It's also fast -- about 6X real time speed. (Resample was 0.8X).
Thank you!
Rainer
RE: correct downsampling
The perfect choral singing means you need to minimize intermodulation distortions in your downsampling process. One of the good ways - to use a perfect soundcard as Lynx L22, for example, for such purpose, it has very good IMD measurements and will distort your signal less that any other soundcard. You may try different dither types, they are implemented in this soundcard as well. It should be good with SSRC and I believe that you will be satisfied finally with the result.
The perfect choral singing means you need to minimize intermodulation distortions in your downsampling process. One of the good ways - to use a perfect soundcard as Lynx L22, for example, for such purpose, it has very good IMD measurements and will distort your signal less that any other soundcard. You may try different dither types, they are implemented in this soundcard as well. It should be good with SSRC and I believe that you will be satisfied finally with the result.
I don't think the problem lies in my sound cards. I hear the same clouding/sharpness on my M-Audio card and my Sony ES CD player when working from the 44/16 version; this distortion is almost totally absent with the 96/24 original through the very same M-Audio card.
True, it gets (much) worse when played on a crummy CD player or a bad sound card. But if the 44/16 version shows problems that the 96/24 version does not, on the very same hardware, it certainly seems like the porblem lies in the resampling -- or in the inherent limitations of 44/16.
The sharpness introduced by ssrc was not obvious when I sampled the original live performance audio at 44/24 instead of 96/24 -- so that sort of rules out the sampling rate. There were other shortcomings with 44/24 sampling (mostly loss of "palbable presence").
This work was done with M-Audio Audiphile 2496 and with M-Audio Delta 1010-LT sound cards.
Rainer
True, it gets (much) worse when played on a crummy CD player or a bad sound card. But if the 44/16 version shows problems that the 96/24 version does not, on the very same hardware, it certainly seems like the porblem lies in the resampling -- or in the inherent limitations of 44/16.
The sharpness introduced by ssrc was not obvious when I sampled the original live performance audio at 44/24 instead of 96/24 -- so that sort of rules out the sampling rate. There were other shortcomings with 44/24 sampling (mostly loss of "palbable presence").
This work was done with M-Audio Audiphile 2496 and with M-Audio Delta 1010-LT sound cards.
Rainer
The problem is not in your sound card, it is in downsampling itself. It is very duifficult to perform it flawlessly. Re-sampling means using of DSP and it's algorithms, and here the card's performance plays very serious role ...Rainervs said:
Correct, you got the distortions at the stage of downsampling and then at the stage of downconversion.
Rainer, Envy 24 DSP chipset never produced the "correct" sound, at least at the level you require.
http://www.io.com/~kazushi/audiocard/audiophile/
Speaking of intermodulation in choral music, read Dave Griesinger's
paper on acoustic intermod created by such choirs.
But, yes, a capella choral music, especially in a reverberant acoustic
is very revealing of processing. I know it most readily shows
up the artifacts in compression like MP3, so I'm not surprised
that you're hearing problems from downsampling.
paper on acoustic intermod created by such choirs.
But, yes, a capella choral music, especially in a reverberant acoustic
is very revealing of processing. I know it most readily shows
up the artifacts in compression like MP3, so I'm not surprised
that you're hearing problems from downsampling.
A quick Google search suggests that the Fraunhofer-Gesellschaft Codec, and the AudioActive Proiduction Studio, are for creating MP3 files. I have almost no interest in MP3. I'm just looking for a way to convert a PCM wave file at 96/24 to a PCM wave file at 44.1/16 and 48/16. Adding the compression problems of MP3 can't possibly help. Any other suggestions?Frazzled said:
Thanks,
Rainer
Thanks for the suggestion. I found his home page, and an incredible bibliography. But I can't recognize the paper you're referring to. And Google can't find anything on his site using the word "choral". Can you give me a more specific reference, or just the correct title?BrianL said:
Thanks,
Rainer
I still don't understand why you're so focused on my sound card. The sound card has NOTHING to do with the resampling process! The resampling is done by a program that reads one WAVe file and writes another WAVe file. If I want to listen to the file, I'll listen at 96/24. But I have to distribute the sound file to people who don't support 96/24. Hence the need to resample into a format that others can use.Gordon McGregor said:
Rainer
RE: resampling
http://www.lynxstudio.com/products.html
Here are the magic words:Rainervs said:
http://www.lynxstudio.com/products.html
It means you can do resampling in hardware ... another alternative, may be even better - to use dCS972 D/D converter ...
Dear Mr. McGregor,
You seem to be focused on a HARDWARE-based solution to the resampling problem. No, I don't have $5,000 to spend for a dSC 972 D/D converter. Nor do I need its broad capabilities.
Except for the approach of using a D/A converter feeding an A/D converter, all the hardware solutions -- whether based on the 972 or on a Sound Blaster -- are simply calculations performed in dedicated silicon. They could equally well be carried out by a general purpose computer. There is nothing to be gained by building them into special purpose chips (except perhaps speed); the result is the same either way.
Which brings me back to my original question: What's a GOOD algorithm for converting 96 kHz audio to 44.1 kHz audio? And what's an accessible, good implementation of that algorithm?
If you have a constructive answer to that question, I look forward to hearing from you again.
Rainer
You seem to be focused on a HARDWARE-based solution to the resampling problem. No, I don't have $5,000 to spend for a dSC 972 D/D converter. Nor do I need its broad capabilities.
Except for the approach of using a D/A converter feeding an A/D converter, all the hardware solutions -- whether based on the 972 or on a Sound Blaster -- are simply calculations performed in dedicated silicon. They could equally well be carried out by a general purpose computer. There is nothing to be gained by building them into special purpose chips (except perhaps speed); the result is the same either way.
Which brings me back to my original question: What's a GOOD algorithm for converting 96 kHz audio to 44.1 kHz audio? And what's an accessible, good implementation of that algorithm?
If you have a constructive answer to that question, I look forward to hearing from you again.
Rainer
Can't you just borrow it or rent the stuff for that particular task? You want to get the best possible level of conversion, right?Rainervs said:
No, the solution works in the digital domain, i.e without D/A or A/D conversion, just using DSP. The same or almost the same DSP can be created in software, you are right.
OK, let's be focused on comparison of algorithms (they are listed in the article):
http://audio.rightmark.org/lukin/dither/dither.htm
It looks like Extrabit Mastering Processor 2.5 is what you need:
http://audio.rightmark.org/lukin/dither/
In theory yes, in practice people, who sell pro equipment, make it more carefully as they can't just fix "a bug" ...
Hopefully I had answered to your question? If no, please accept my apologies, I will not bother you anymore.
Gordon McGregor said:
Extrabit only seems to do word length changing (well thats all the web site talks about)... thats not really too important (its not difficult to cut off 8bits and dither).
The difficult part is the 96Khz to 44.1/48Khz resampling.
I feel for you Rainer... this shouldnt be that difficult
I would just use Cooledit Pro, but im not sure if it would be better than the other methods you have tried.
Maybe download the demo of Cooledit Pro and try it?
Thank you for the concrete suggestion, Gordon.Gordon McGregor said:
Unfortunately, Extrabit does not do sample rate conversion; it only dies bit depth conversion (like 24-bit to 16-bit). It's all about dither. And, after spending an hour or two studying dither choices (http://www.24-96.net/dither is a good place to start), I conclude that dither is not my issue; it's SAMPLE RATE conversion (96 kHz -> 44.1 kHz).
Apology accepted.
Rainer
OK, try foobar2000 for the whole conversion/downsampling, this is a kernel-streaming software http://209.152.181.169/foobar2000/FAQ.htmlRainervs said:
Yo can also try to do downconversion in Extrabit and then downcsampling in foobar ...
Well, maybe this is why so many CDs have sounded so bad over the last 15 years ...MWP said:
I feel for you Rainer... this shouldnt be that difficult
I would just use Cooledit Pro, but im not sure if it would be better than the other methods you have tried.
Maybe download the demo of Cooledit Pro and try it?
Cooledit Pro is a good suggestion, but (fortunately or unfortunately) last summer Syntrillium sold the whole Cooledit family to Adobe, where is has been relabelled as Adobe Audition 1.0. All the major download sources for Cooledit just point to an Adobe announcement page. And, no, Adobe doesn't seem to much believe in demo versions.
On the other hand, Adobe Auditon 1.5 has some intriguing frequency-domain editing capabilities (good for removing coughs, door closings, organ blower noise, etc.) that might be helpful. I just hate to invest $250 when I don't know what I'm getting.
Rainer
What are you doing, looking for hits in Google and posting random results to me? I'm getting tired of this runaround from you!Gordon McGregor said:
FooBar2000 is a MUSIC PLAYER, not a sample rate converter. Yes, it offers to do UPsampling as part of the playing process, but that is not its reason for being. The web page description does not talk about a DOWNsampling carability at all; and there is no hint that it is willing to save its resampled output to a file. So it really sounds like it isn't at all what I'm looking for.
No, you haven't answered my question. Yes, I've accepted your apology. Now I expect you to stick to your promise to stop bothering me (unless you have a REAL answer from FIRST-HAND experience).
Rainer
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.