How can samplers claim to produce 192kHz sound?

[dn83]

How can samplers claim to produce 192kHz sound when they use 44.1kHz samples?

I might be missing something here, but how can samplers such as Kontakt or DirectWave (and even Sampler, I guess) claim to output at true 192kHz (or 96kHz, 88.2, 48 or whatever sample rate you're using for your project) if they use 44.1kHz .wav sound files? (You can see this if you check out the file properties of any .wav in the Kontakt or DirectWave library.) Are they UP-sampling to sample rates above 44.1kHz?

[nebulae]

The actual sample is the basic waveform for creating sound. The sound is then interpollated, filtered, run through effects, etc. The output of that sound is then at the samplerate of your project. This has no bearing on sound quality, simply on the output of the synth. If it's a crappy sample, it'll still sound crappy in the output.

[dn83]

Hey, thanks for the reply. Ah yeah, I was suspecting something like that.

So by interpolation you mean it basically rebuilds the sample's waveform to the needed samplerate by inserting predicted data for the missing data?

[nebulae]

I believe interpollation has to do with how a sample gets pitched as you play it across a keyboard. Meaning if the root key is C3, then when you play D3, the sample has to be interpollated.

[dn83]

Hmm I see. Well I guess the difference in sound quality is trivial really, from 44.1kHz to 48kHz (what I use in projects) then back to 44.1kHz for an mp3, for example. But I just assumed for example, that Kontakt has true 192kHz samples of say, a Grand Piano then down-samples to your project's sample rate. But I guess the files would be HUGE, haha.

I'm just a stickler for sound quality - I don't like to sample then resample then resample, especially if the sample rate of the original sample is low to start with. Oh well, bit rate factors in too (16, 24 or 32) so anyway...interesting as always!

[nebulae]

I hear no real difference between 44.1 and 48. I stick with 44.1 simply because it's the CD standard, so there's resampling later. Also, most samples and loops are 44.1 so instead of having pitch issues, I stick with 44.1.

Most people I know stick with 44.1, 24-bit. Best balance of quality, space, and CPU usage.

[Tone Deft]

There's less loss downsampling than upsampling, less interpolation.

Interpolation means to find a new data point based on data points you know
http://en.wikipedia.org/wiki/Interpolation

So by interpolation you mean it basically rebuilds the sample's waveform to the needed samplerate by inserting predicted data for the missing data?

Pretty much, the process is sample rate conversion, interpolation is used in that process, it's like saying multiplication is used. You don't say you multiply a signal, you say you used multiplication.

Oh well, bit rate factors in too (16, 24 or 32) so anyway...interesting as always!

Bit rate is the sample rate times the bit depth, the 16, 24 and 32 numbers you refer to are the bit depth. At 44.1kHz, 16 bits, you have 705,600 bits every second, or 705.6kbps. The bit depth is how you get the dynamic range, 2^16 means you have 65,536 numbers in those 16 1s and 0s. Converted to decibels that's 96.3dB

Some good wikis to read (these are short and to the point, no math in the first 2)
http://en.wikipedia.org/wiki/Dynamic_range
http://en.wikipedia.org/wiki/Bit_resolution
http://en.wikipedia.org/wiki/Decibel <- some math but very good to know

[dn83]

Hey guys, thanks for the replies, good tips! Ooops, yeah I meant bit depth, thanks for setting that straight. I'll check out the wiki articles to brush up on this stuff.

So samplers like Kontakt and DirectWave can technically output at 192kHz 32 bit, but unless the original sample (.wav or .aif) files used are also 192kHz 32 bit, then it's not truly 192kHz 32 bit, eh?

Otherwise it's just a 44.1kHz 16 bit sample being artificially represented as being 192kHz 32 bit? But any modification/effects added by the sampler are truly 192kHz 32 bit?

Oh well, sounds fine to me, I was just interested in why a sampler might be 192kHz 32 bit capable yet uses 44.1kHz 16 or 24 bit samples, haha. I do realize though, that 192kHz 32 bit files are humongous! LOL At what point do you reach diminishing returns with sample rates and bit depths? Does it really just depend on how many times a sound will be resampled?


cheers,
Doni

[Michael-SW]

I hear no real difference between 44.1 and 48. I stick with 44.1 simply because it's the CD standard, so there's resampling later. Also, most samples and loops are 44.1 so instead of having pitch issues, I stick with 44.1.

Most people I know stick with 44.1, 24-bit. Best balance of quality, space, and CPU usage.

Amen to that. Working in 48 kHz will probably not net you anything (except lots of sample rate conversions) unless you work extensively with DAT recordings. 48 kHz was a sample rate invented so that it wouldn't be easy to copy digital audio from CDs to DAT, back in the days when DAT was expected to be a big consumer product.

Working in 24 bits is almost essential on the other hand.

[dn83]

Ahh, I see. Thanks again, good tip - I suppose I agree. 48kHz yields not much difference, but I guess I was thinking of allowing some padding for final mix-down. But again, like you guys say - probably not worth it.

While we're on the topic, why does the supported sample rates (at least for my audio interface) jump from 44.1 to 48 then all the way to 88.2kHz, with nothing in-between?

Also, why is a bit depth of 24 (for initial samples) more common than 32?

[Angstrom]

the reason a lot of softsynths work at a higher internal fequency than you can actually hear is simply so they get the maths right on their transforms. Some effects such as distortions or even filters require high bandwidth so they don't alias down unwanted frequencies.

putting it simply!

it's just so they can calculate it better. 44.1 is fine enough for output, but for calculation of that output you would be suprised at the frequency headroom which is sometimes required. That's whats often referred to as 'oversampling', you will see 4x or even 8x oversampling in some filters for example.
Upsampling a crappy input may actually benefit if you are applying a lot of processing, in some ways you can actually polish that turd.

you still have a shiny turd at the end though, but perhaps that's what you want?

[Tone Deft]

why does the supported sample rates (at least for my audio interface) jump from 44.1 to 48 then all the way to 88.2kHz, with nothing in-between?

Those are all the accepted standard sampling rates, 32k, 44.1k, 48k, 88.2k, 96k and 192k (I might be missing one). Note that 32k is only good for sampling the human voice as you can only capture signals up to 16kHz (FM radio quality.)

Also, why is a bit depth of 24 (for initial samples) more common than 32?

I believe it has to do with the clocking and bit coordination. 32 bits of data don't leave room for overhead bits in the data stream.
Internally to audio equipment, there's word clock, bit clock and serial data.
Word clock runs at the sample rate (Fs = sampling frequency), it's a square wave, 1 and 0.
The serial data runs at the bit clock rate. When the audio ICs see a rising edge on bit clock, they sample the data stream for the data. The data is shifted so that the rising edge of bit clock is right in the middle of the 1 or 0 in the data stream.
When word clock is 1, the data stream is the left channel, when it's 0, you're seeing right channel.
Bit clock runs at 64*Fs, 128*Fs or 256*Fs, splitting word clock into 64, 128 or 256 slots, each slot is where a bit can be located.
At 64*Fs you have left and right channel, 32 bits each, 24 bits of data, 8 bits of user data (not always used).
128*Fs you have twice as many channels, 4 channels of 32 bit
At 256*Fs you can carry 8 channels of data. I worked on a project where we squeezed in 10 channels but that was kinda cheating. Likewise you can fit 32 bits of audio into 64*Fs if you don't use the overhead bits and your ICs don't step on it.

Dunno, if that helps, it's more like that's the way it is rather than why it is, call it necessity. I'll think about it more if I missed something, I never thought about why, I just do it.

This might help
http://en.wikipedia.org/wiki/I%C2%B2C

Also, SPDIF/AES/EBU is very similar except it's one signal, data combined with a clock in a way that the clock is recoverable automatically.

[ethios4]

thanks TD, that was nice!

I believe the sample rate jump goes from 44.1 and 48 to 88.2 and 96 because the higher rates are double the lower rates, which makes for cleaner interpolation for upsampling? 4x48=192...is 192 currently being pushed for the film industry?

[Tone Deft]

Film mastering and playback uses 24 bit 48kHz. Dolby's AC3 and DTS encoded material are sent as DATA bits rather than audio bits, a flag is set in the AES/EBU (SPDIF) stream that tells equipment not to decode the data, once it hits a DSP and is decoded THEN that flag is returned to normal. Straight PCM is always PCM, you can feed that straight into a D/A converter.

[willb]

So samplers like Kontakt and DirectWave can technically output at 192kHz 32 bit, but unless the original sample (.wav or .aif) files used are also 192kHz 32 bit, then it's not truly 192kHz 32 bit, eh?

If someone is selling 192khz samples, you should ask to see the microphone they used. I'm no expert, but I don't know of any microphone with any frequency response in the 96khz range (the maximum representable frequency of a 192khz file). For comparison, a standard prosumer-grade LDC can pick up frequencies up to about 18khz, which would only require a 36khz sample rate to reproduce.