How to downconvert 96 kHz audio to 44.1 Khz for CD burning?

[bryantm3]

Hi,
I've had a lot of help here before, but I wasn't sure which section to use, so I went here. Anyway, I've taken a bunch of old vinyl recordings and recorded them onto my computer at 96 kHz/16 bit WAV files, so that I can remove pops and clicks more cleanly and that I don't lose any of the sound when I do this. So now I need to downconvert these 96 kHz files to 44.1 kHz so I can burn them onto a CD, because when I burn them straight onto a CD, it sounds a bit funny. I don't know what application to use, but I'd like to do it so that it doesn't lose a lot, like I'm not sure if I want to do it in iTunes. So if anyone has any suggestions for a free application to downconvert 96 kHz to 44.1 kHz, please tell me. By the way, I'm using a Mac, so if that limits my options, I'm not sure. Any help would be appreciated.
Thanks,
Dan

 

[MDS]

I don't know about any free applications but any auditor that supports 96 kHz sample rates can downsample to 44.1 kHz.

What application did you use to record at 96 kHz and what sound card do you have?

Burning 96 kHz files to CD shouldn't just sound funny - it shouldn't work at all - so whatever app was used to burn the CD must have downsampled on the fly.

 

[bryantm3]

I used Audacity to record at 96 kHz, and I have the standard sound card that comes with the 2005 iMac, which can support up to 96 kHz/24-bit. What do you mean 'Auditor'?

 

[MDS]

Sorry it was a typo. I meant 'audio editor' - like Sound Forge, Adobe Audition, ProTools, etc. I guess my brain combined audio and editor into one word.

Can you verify that the recorded file is actually 24/96? A 3 minute WAV at 24/96 would be over 103 MB.

I'm not familiar with the default sound card that comes with a Mac but I can tell you that many sound cards that claim to support 24/96 or higher don't really support it. It depends on the underlying audio processing chip in the sound card. For example, Creative cards claim 24/192 but in reality the max it supports is 16/48 because that is the native format the chip uses. If the app tells the sound card to use 24/96, what it really does is sample at 16/48 and then zero extend the 16 bit samples to 24 bits and then double each sample so you now have 24/96. There is no quality difference from simply recording at 16/48 in the first place.

Likewise, if the app tells the sound card to record at 16/44.1, what it really does is sample at 16/48 then downsample to 16/44.1 on the fly.

Edit: If you intend to burn the files to CD, you should record at 16/44.1 unless you know for a fact the native sampling rate and bit depth of the sound card's audio engine. I know my Audigy 2 ZS is really 16/48 so I record at 16/48 using SoundForge and then downsample to 44.1. That avoids having the sound card do the downsampling on the fly and I have SF do the downsampling afterwards.

 

[WmAx]

For example, Creative cards claim 24/192 but in reality the max it supports is 16/48 because that is the native format the chip uses. If the app tells the sound card to use 24/96, what it really does is sample at 16/48 and then zero extend the 16 bit samples to 24 bits and then double each sample so you now have 24/96.

The only Creative card of which I know about for which this was true was the Soundblaster Audigy 1. In later cards, 24/96 was supported. In the newest line, Xi-Fi, up to 192/24 is supported. However, if you enable the EAX sound F/X on most of the cards, regardless of possible maximum support, the card will work at 48kHz sample rate, because the signal will be routed through a CPU process intended for game play only.

-Chris

 

[MDS]

Nope, only the latest XFi cards. I forget the name of the chip the Audigy line uses but it is the same 'ol same 'ol - 16/48.

They support playback of 24/96 and 24/192 by downsampling to 16/48 and they support 'recording' of 24/96 by playing the game I described above.

 

[bryantm3]

It's not 96/24, it's 96/16. Each file is around 6 minutes long and 140 MB each. I zoomed in on the wave, and each sample is at a different amplitude than the one before it, so it's definitely really 96 kHz.

but anyway, my point is, how do i get these downsampled to 44.1 kHz in Audacity or some other app?

 

[MDS]

You have Audacity...so see if there is a menu option for downsampling.

If it does there should be a few parameters:
- Interpolation Accuracy: set it to the highest unless you are so impatient that you want it done in one minute instead of three.
- Antialias filter: yes

I don't mean to belabor the point but just because there are 96K samples per second it does not necessarily mean that it was actually sampled at 96 kHz. Only expensive 'pro' sound cards can actually sample at that rate and all the others twist the truth via marketing.

The question of whether to sample at high sampling rates and bit depths when the final destination is a CD comes up often at the Sound Forge forums and the answer from the recording engineers is ALWAYS sample at 16/44.1 if you intend to burn a CD. Consumer level sound cards play the game I described above and all you get is a much larger file that contains the exact same information as if it had been sampled at the card's native bit depth and sampling rate.

 

[WmAx]

Nope, only the latest XFi cards. I forget the name of the chip the Audigy line uses but it is the same 'ol same 'ol - 16/48.

They support playback of 24/96 and 24/192 by downsampling to 16/48 and they support 'recording' of 24/96 by playing the game I described above.

I have had these cards. I measured them. What you seem to be claiming (that non Xi-Fi Creative cards can't play or record actually beyond 48/16) is incorrect outside of the exceptions I stated in the prior post.

-Chris

 

[bryantm3]

Yeah, the only reason I sampled that high is because I was removing pops and clicks and this program I used will only delete the clicks, and I thought that it would be more accurate with a higher sampling rate.
But anyway, as I said before, I'm pretty sure it's really 96/16 because the samples don't repeat themselves like they would if it was just oversampled.

 

[WmAx]

I don't mean to belabor the point but just because there are 96K samples per second it does not necessarily mean that it was actually sampled at 96 kHz. Only expensive 'pro' sound cards can actually sample at that rate and all the others twist the truth via marketing.

I have one card, a Creative Soundblaster Live! 24 bit. A $30 PCI card 2 years ago when it was still on the market. 96/24 recording mode, measured, has about a 101 dB SNR and has basically a flat response up to a little past 40,000 Hz.

Not an expensive pro card by any means.

-Chris

 

[MDS]

Sorry WmAx but you are incorrect. Creative cards have been criticised for years precisely because of these issues. You may put it in 24/96 mode and measure it but it is NOT truely playing back in that mode. A 24/96 input will be resampled to 16/48, processed, and then upsampled back to 24/96. Your tools tell you that the output is 24/96 but it is not the same 24/96 that went in.

It is not magic. The underlying audio processor has a native bit depth of 16 and native sampling rate of 48 kHz. That is all it can do, no ifs ands or buts about it. It will accept 24/96 or 24/192 but the first thing it does is resample to 16/48 because the processor only works at that rate. This is conceptually no different than a TV that accepts 1080p but has a native resolution less than that and converts the incoming signal to its native resolution.

When you *record* at say 24/96 what it does is bit depth conversion from 24 to 16 (which is a lossy conversion) and resamples to 48 kHz (which basically throws away every other sample). Then it zero extends the 16 bits to 24 which of course adds no information because adding leading zeros doesn't change the value. It then dups (or interpolates) the 48K samples to get 96K samples and the result is not true to the original. To the *ear* it will sound fine but it is not the same.

Why do you think the 'fanatics' recommend M-Audio and Edirol cards? Because they do not convert the incoming bit depth and/or sample rate. If you specify 16/44.1 that is what you get. It's not the case with Creative cards.

As I stated, that is why I record at 16/48 so the card does not resample for me. I do the resampling after the fact with Sound Forge.

 

[WmAx]

Sorry WmAx but you are incorrect. Creative cards have been criticised for years precisely because of these issues. You may put it in 24/96 mode and measure it but it is NOT truely playing back in that mode. A 24/96 input will be resampled to 16/48, processed, and then upsampled back to 24/96. Your tools tell you that the output is 24/96 but it is not the same 24/96 that went in.

The Creative cards which claim 24/96 ( excepting the Audigy 1 which always resamples ) only resample 96/24 to 48/16 when using the EAX F/X CPU on the card, or in the case of some, if you manually set the re-sample mode. The processor to which you refer is the EAX engine only, though I forget the actual name of the chip. It is bypassed unless you expressly activate EAX F/X.

As for my measurements, they tell me what the card actually does. It records and plays back a bit depth exceeding 16 bits, and a sample rate exceeding 48kHz. There is no re-sample algorithm that can re-create non-existant information later. The process of down sample reduces bandwidth. This is simply not the case here, unless you activate the EAX F/X CPU.

As for audiophiles recommending whatever cards, who cares? Most audiophiles appear to be ignorant by default.

-Chris

 

[MDS]

Creative cards always resample. The underlying audio engine only works at 16/48. Period. Google is your friend or pop into the Sound Forge forums and ask there as the professional recording engineers there will give you the straight scoop.

 

[WmAx]

Creative cards always resample. The underlying audio engine only works at 16/48. Period. Google is your friend or pop into the Sound Forge forums and ask there as the professional recording engineers there will give you the straight scoop.

What good is asking someone that is spreading misinformation about said subject? It matters not what label is affixed to said person(s); facts will not change simply because a particular person claims otherwise. I actually bothered to measure the characteristics of the analog I/Os, unlike the person(s) making the ignorant claims, whom is/are apparently guessing functionality, if your representation is accurate.

If you do not enable EAX F/X(E.Q., reverb, etc.), the Creative 96/24 Cards (except Audigy 1) will record at up to 96/24 through the analog inputs, and playback up to 96/24 through the analog outputs. There is no apparent evidence that the cards re-sample 96/24 data unless you enable the EAX or manually select another sample rate (an option on some cards). There appears to be a provision to bypass the re-sample engine that is the center of the EAX F/X in order to allow proper 96/24 recording/playback. The digital outputs/inputs may have some kind of default re-sampling, but I have no idea, as I have not measured the digital outputs/inputs.

-Chris

 

[tbewick]

The question of whether to sample at high sampling rates and bit depths when the final destination is a CD comes up often at the Sound Forge forums and the answer from the recording engineers is ALWAYS sample at 16/44.1 if you intend to burn a CD. Consumer level sound cards play the game I described above and all you get is a much larger file that contains the exact same information as if it had been sampled at the card's native bit depth and sampling rate.

I'm under the impression, along with Chris, that most modern sound cards do allow for 24 bit wordlengths where the 8 additional bits over 16 bit operation does contain actual audio information. This set of measurements for the Creative Audigy 4 Pro show that the card appears to have best performance operating in 32 bit mode and at 192 kHz:

http://www.trustedreviews.com/multimedia/review/2005/06/14/Creative-Audigy-4-Pro/p2
Paul Weir, TrustedReviews, 14 June 2005.

The dynamic range of the card is 96 dBA in 32 bit/192 kHz mode, and falls to 86 dBA in 16 bit/48 kHz mode. I haven't bothered reading about how the test was conducted but I'm guessing that the difference in the results is a real one.

bryantm3,

My own recommendation would be to do the recording at the highest sampling rate and using the greatest number of bits available on the card. It is likely that with most cards this will produce the best results. The recorded file can then be truncated to 16 bits with TPDF-dither which can be carried out with very high precision in the digital domain. You could also choose to noise-shape the added quantization noise to improve the subjective dynamic range.

I would also say that the filtering needed to reduce a recording made with a sample rate higher than 44.1 kHz to 44.1 kHz will probably produce better results than simply recording at 44.1 kHz. One thing is that the signal-to-noise of the recording will be improved since less noise will exist inside the audio band. This extra accuracy of the recording will improve the results of the finished downsampling.

 

[MDS]

What good is asking someone that is spreading misinformation about said subject? It matters not what label is affixed to said person(s); facts will not change simply because a particular person claims otherwise. I actually bothered to measure the characteristics of the analog I/Os, unlike the person(s) making the ignorant claims, whom is/are apparently guessing functionality, if your representation is accurate.

So other professionals are spreading misinformation because their results do not agree with yours? You are in the minority with your beliefs.

 

[MDS]

My own recommendation would be to do the recording at the highest sampling rate and using the greatest number of bits available on the card. It is likely that with most cards this will produce the best results. The recorded file can then be truncated to 16 bits with TPDF-dither which can be carried out with very high precision in the digital domain. You could also choose to noise-shape the added quantization noise to improve the subjective dynamic range.

That would be a great recommendation if the card in use actually records at the higher sampling rate and bit depth. Creative cards up to and including the Audigy line do not.

You should record at the card's native resolution and then do the bit depth conversion and resample after the fact. For Creative cards that is 16/48.

 

[WmAx]

So other professionals are spreading misinformation because their results do not agree with yours? You are in the minority with your beliefs.

[sarcasm]Right. Well, Creative should get an award then, for making the first up sample algorithms that actually restore non-existent recorded bandwidth that was *never* sampled.[/sarcasm]

-Chris

 

[WmAx]

That would be a great recommendation if the card in use actually records at the higher sampling rate and bit depth. Creative cards up to and including the Audigy line do not.

You should record at the card's native resolution and then do the bit depth conversion and resample after the fact. For Creative cards that is 16/48.

From a 3rd party measured review:

"A frequency response will also be a useless parameter if you don't account for the fact that a manufacturer can get it wrong. The Audigy, in spite of 24bit 96kHz supported, couldn't produce the whole frequency range in this format accurately, - it converted it into 16bit 48kHz and had a bad cut in the HF audible range (attention! it's only for the 24/96 mode).

In the Audigy2 the conversion is removed and the frequency response is corrected. It would be better if Creative gave in specs the frequency response for the audible range. Luckily, it's possible with the RMAA program. Well, the Audigy2 can produce the whole range in the 24/96 mode without even a hint of any conversion and with a remarkable FR similar to that of professional cards. "

http://www.digit-life.com/articles2/creativeaudigy2/index.html

-Chris