Kemper Profiler MK 2

[Orvillain]

Points aren't used in FFT's or STFT's guys. They're called bins.

Personally I do not see any reason why this sentence would refer to FFT at all:

• More than 100,000 individual frequency points meticulously analyzed for the most powerful amp recreation ever achieved.

This is just marketing bollocks at the end of the day. It doesn't fucking mean anything.

 

[Orvillain]

Also, there are serious trade-off's with a high number of FFT bins (IE: high frequency resolution)

Generally speaking, the higher the frequency resolution, the lower the time resolution. You get temporal smearing artifacts.

It is an unavoidable facet of FFT and STFT processing.

You can also get leakage from one bin to another, depending on how you've configured your FFT window. EVen if you use a Blackman-Harris window to minimize the issue, it is always going to be present on some level.

You can do multi-resolution approaches, but they're expensive.

You can also weight the bins towards frequency bands that are more musically meaningful; this will enable you to reduce the bin count, while still processing the range of frequencies that you care about.


I recently wrote an FFT based processor to extract spectral features from audio, and this is a delicate balancing act and is very material dependent.

 

[dws2384]

Also, there are serious trade-off's with a high number of FFT bins (IE: high frequency resolution)

Generally speaking, the higher the frequency resolution, the lower the time resolution. You get temporal smearing artifacts.

It is an unavoidable facet of FFT and STFT processing.

You can also get leakage from one bin to another, depending on how you've configured your FFT window. EVen if you use a Blackman-Harris window to minimize the issue, it is always going to be present on some level.

You can do multi-resolution approaches, but they're expensive.

You can also weight the bins towards frequency bands that are more musically meaningful; this will enable you to reduce the bin count, while still processing the range of frequencies that you care about.


I recently wrote an FFT based processor to extract spectral features from audio, and this is a delicate balancing act and is very material dependent.

So what’s your wild guess on what 100k points means then?

It must be correlated to something similar to what they’re already doing since the mk2 captures will still play on Mk1 at lower resolution. To me that suggests the new process is similar but just has more data points?

 

[Orvillain]

So what’s your wild guess on what 100k points means then?

It must be correlated to something similar to what they’re already doing since the mk2 captures will still play on Mk1 at lower resolution. To me that suggests the new process is similar but just has more data points?

If I was to be ultra cynical, I'd say it could be some kind of summation of various areas or levels of detail, but I really don't know.

Ultimately, it really doesn't mean very much as a sentence.

One golden rule of marketing - you see big number, you mistrust big number. Someone put it there for a reason; to make you think a certain way.

 

[LP59]

So what’s your wild guess on what 100k points means then?

It must be correlated to something similar to what they’re already doing since the mk2 captures will still play on Mk1 at lower resolution.

This is a poorly worded marketing blurb. There is no "must be correlated" about it. You're taking a microscope to something that was never meant to have any meaning in the first place.

 

[Chocol8]

So what’s your wild guess on what 100k points means then?

Well, it seems to be intentionally deceptive, so my not so wild guess is that it means the person behind that statement is not the worlds most honest salesman.

 

[dws2384]

Ultimately, it really doesn't mean very much as a sentence.

This is a poorly worded marketing blurb. There is no "must be correlated" about it. You're taking a microscope to something that was never meant to have any meaning in the first place.

Well, it seems to be intentionally deceptive, so my not so wild guess is that it means the person behind that statement is not the worlds most honest salesman.

Sounds promising! Can’t wait!

 

[Jarick]

Could be 100k spread across multiple measurements that are used to train the AI model and not the output file.

 

[OneEng]

FFTs don’t inherently require powers of two—radix-2 FFTs do for maximal speed
Mixed-radix and Bluestein’s algorithms let you handle arbitrary lengths like 1000 with only a small performance overhead.

True, and thanks for the clarification. I guess I have just been doing it for speed's sake. Certainly the one I was forced to write in college only worked with radix of 2. The ones I have used in libraries also prefer it, but as you say today's FFT libraries can handle other lengths.

Could be 100k spread across multiple measurements that are used to train the AI model and not the output file.

It's hard to say. I suspect (but don't know) the following:

1) The new algorithm will show improvement over the old. It may reach near parity with NAM.
2) It will likely be done predominantly on the PC.
3) I am hoping that they rid themselves of the "refinement" steps and make it more automated.

 

[TubeStack]

I can’t wait to make a Profile of a Mk2 Kemper Profile using my Mk1 Kemper and it sounds IDENTICAL to the Mk2 Profile!