SRC and Dither question.....

[captain caveman]

Hi

I am wondering what the best procedure is for SRC and dithering.

If I have a project at 48kHz 32 bit float, is it best to export a WAV at 48kHz 16bit then do the SRC, or is it OK to export at 48kHz 32bit with the dither and then do the SRC and bit-depth change at the one time?

I've always just used the SRC and bit depth reduction in Cubase, but I have read that Voxengo's R8Brain is far superior so I will be using this as a seperate process, hence the question.

Appreciate any info.

[16/44.1]

Originally Posted by captain caveman

Hi

I am wondering what the best procedure is for SRC and dithering.

If I have a project at 48kHz 32 bit float, is it best to export a WAV at 48kHz 16bit then do the SRC, or is it OK to export at 48kHz 32bit with the dither and then do the SRC and bit-depth change at the one time?

I've always just used the SRC and bit depth reduction in Cubase, but I have read that Voxengo's R8Brain is far superior so I will be using this as a seperate process, hence the question.

Appreciate any info.

From 48kHz/32bit to 48kHz/32bit, there's no need to dither, it's the same.
Perhaps you mean from 48kHz/32bit to 44.1kHz/32bit?
If so, keep it 32bit during SRC.
Only convert it to 16bit (with dither in that case) after SRC and then go to a medium which is capable for 16bit.

[captain caveman]

Sorry, I read that again and I didn't make sense the first time.

I am wanting to go from 48kHz 32bit to 44.1kHz 16bit for CD.

So the dither should be added after the SRC down to 44.1kHz as a seperate process following putting mastering effects on at 48kHz 32bit?

Would saving a 48/32 .wav with mastering effects and dither and then getting R8Brain to do an SRC and bit depth reduction be worse than the first way?

Please excuse the nit-picking, but I want to get the most out of my relatively humble set-up and I like to cross the Ts and dot the Is.

[Bob Olhsson]

It really comes down to what is your best sounding sample rate converter and what is your best sounding dithering and bit reduction tool.

You just need to do your homework and listen very carefully to all of your options.

[16/44.1]

Originally Posted by captain caveman

Sorry, I read that again and I didn't make sense the first time.

I am wanting to go from 48kHz 32bit to 44.1kHz 16bit for CD.

So the dither should be added after the SRC down to 44.1kHz as a seperate process following putting mastering effects on at 48kHz 32bit?

Would saving a 48/32 .wav with mastering effects and dither and then getting R8Brain to do an SRC and bit depth reduction be worse than the first way?

Please excuse the nit-picking, but I want to get the most out of my relatively humble set-up and I like to cross the Ts and dot the Is.

You don't need dither when you go from 48/32 to 44.1/32.
You must always use dither if you want to reduce wordlength.
32bits to 16 bits.

[captain caveman]

Thanks for the replies guys.

I have thought about this and now realise that it doesn't matter which way I do it because the dither noise will be SRCd alongwith the rest of the audio and so the subsequent bit depth reduction will take place in a file with SRCd dither noise.

The only difference would be an slight degradation of the dither noise due to the SRC but if I am using an SRC of the quality of R8brain this will not matter.

Apologies for the user head gap.

[captain caveman]

Although in an ideal world the best way would be to go from the higher sample rate and bit depth with effects down to 44.1kHz 32 bit and then dither down to 16 bit.

But then SRCd noise is still just noise - right?

A bien tot mes amis.

[UnderTow]

I would personaly SRC and then reduce bit depth or just use R8Brain as it dithers when bit reducing so you have SRC and bit reduction in one easy step.

Originally Posted by captain caveman

Thanks for the replies guys.
The only difference would be an slight degradation of the dither noise due to the SRC but if I am using an SRC of the quality of R8brain this will not matter.

This is irrelelvant. The dither noise is only "important" at the time that you reduce the bit depth. After that it is just part of the noise floor of your audio. It has no more effect except raising the noise floor slightly.

Alistair

[Bob Olhsson]

Sample rate conversion math creates large bit depths that need to be reduced in order to store the audio as a standard computer file. Dither makes the bit reduction sound like noise rather than like distortion. Dither is not just noise that covers up the distortion so you need to dither every time you reduce bit depth in order to avoid distortion.

The least damaging way to do this is by taking a 24, 32 or 64 bit file, converting the sample rate and then reducing the number of bits to your target bit-depth using dither. This can all be done in one step or the conversion and final reduction can be done separately by making a 24/32/64 bit intermediate file when your sample rate converter's bit reduction doesn't sound as good as that of your other audio applications.

[captain caveman]

Thanks for taking the time out for that concise reply Bob, much appreciated.

Also thanks UnderTow for the heads up on R8Brain doing dither too - I didn't know that.

I'm good to go now.

[Andy Krehm]

Originally Posted by Bob Olhsson

Sample rate conversion math creates large bit depths that need to be reduced in order to store the audio as a standard computer file. Dither makes the bit reduction sound like noise rather than like distortion. Dither is not just noise that covers up the distortion so you need to dither every time you reduce bit depth in order to avoid distortion.

The least damaging way to do this is by taking a 24, 32 or 64 bit file, converting the sample rate and then reducing the number of bits to your target bit-depth using dither. This can all be done in one step or the conversion and final reduction can be done separately by making a 24/32/64 bit intermediate file when your sample rate converter's bit reduction doesn't sound as good as that of your other audio applications.

So just for clarification, you are saying that one can go from 48k to 44.1k or 96k to 44.1k without dithering as long as the bit rate stays the same, i.e., keeping 24 bits with the examples above.

I was told by the owner of a well known, high end company that sells harware and software SRCs that you have to apply dither when sample rate convering and offerred to refer me to some "white papers" on the subject. I just took his word for it but may it's not so cut and dried.

So what is the "truth" here?

[MattGray]

Originally Posted by Andy Krehm

So just for clarification, you are saying that one can go from 48k to 44.1k or 96k to 44.1k without dithering as long as the bit rate stays the same, i.e., keeping 24 bits with the examples above.

I was told by the owner of a well known, high end company that sells harware and software SRCs that you have to apply dither when sample rate convering and offerred to refer me to some "white papers" on the subject. I just took his word for it but may it's not so cut and dried.

So what is the "truth" here?

Hey Andy, there is a couple of 64bit float SRC's available which allow the wordlength to grow with sufficient headroom during the SRC. I guess it really depends on the output file resolution as to whether you would need to dither. If you are outputting to a 24bit file it might be wise to use a type 2 TPDF dither. If you are outputting to a 64bit or 32bit file then I can't see there being any point in dithering. I usually use a 32bit 96kHz file & capture at 32bit 44.1kHz then dither that to the target rate (either 24bit or 16bit). If using Waveburner to compile the disc I usually dither to 24bit so I can do fades etc & then output with 16bit MBIT+ dither on bounce project.

[Bob Olhsson]

Assuming I wanted to use something else for dithering down to 16 bits, I would dither the output of the SRC to 24 bits. If it doesn't offer dither, I'd save my intermediate files at the highest resolution offered, ideally 32 bit fixed or 64 bit float.

Again, do your homework and test. Listening with an open mind is very important. I'd never just assume that a software developer knows what they are doing.

[Andy Krehm]

Originally Posted by Bob Olhsson

Assuming I wanted to use something else for dithering down to 16 bits, I would dither the output of the SRC to 24 bits. If it doesn't offer dither, I'd save my intermediate files at the highest resolution offered, ideally 32 bit fixed or 64 bit float.

Again, do your homework and test. Listening with an open mind is very important. I'd never just assume that a software developer knows what they are doing.

The reason I would find it advantageous to dither separately is b/c I SRC 24/96 or 24/48k files to 44.1 and would like the option to do fades and volume changes in my sequencing program (WaveBurner) where I can use POWR dither when writing the master.

If the processes can be technically done separately, it avoids double dithering in case one needs to do a fade or volume change before writing the master due to a miscalculation in the processing program.

The SRC software (Saracon) I am demoing doesn't allow SRC without dither but I noticed that other SRC's do, such as WaveEditor.

So besides listening (which I'm not discounting), is it technically OK to do the processes separately.