|
 Kyma Forum
 Tips & Techniques
 How many FFT bands can Kyma do?
|
| next newest topic | next oldest topic |
| Author | Topic: How many FFT bands can Kyma do? |
|
UmFuFu Member |
posted 21 March 2007 12:28
 
Hey group, I'm a bit confused by the whole FFT analysis process. Mainly how it seems to hide the amount of bands it is going to perform on your sample by the frequency radio boxes. I'm curious if you know how many bands those boxes correspond to and ultimately how many bands you can get Kyma to do in realtime. [This message has been edited by UmFuFu (edited 21 March 2007).] IP: Logged |
|
SSC Administrator |
posted 22 March 2007 08:40
In the SpectralAnalysis module (or the SpectralAnalysis Tool), you can choose a frequency corresponding to the lowest frequency that can be analyzed. (the FFT module does not have radio buttons for Frequency so I guess you are talking about the SpectralAnalysis modules). 1F corresponds to 512 bands (starting at around 42 hz and spaced 42 hz apart), 2F corresponds to 256 bands, 3F 128 bands, etc. IP: Logged |
|
UmFuFu Member |
posted 22 March 2007 14:53
Awesome, thank you. So then 5F is 512 bands? Is this the most you can get Kyma to do? (does using Loris go around this?)
_____ Oh, sorry, I read over your post too quickly. 1F is 512 bands. [This message has been edited by UmFuFu (edited 22 March 2007).] IP: Logged |
|
SSC Administrator |
posted 22 March 2007 22:01
What's the nature of the sound you are analysing? Does it have a fundamental frequency (or is it noisy)? IP: Logged |
|
UmFuFu Member |
posted 22 March 2007 22:13
I would like to do an FFT/Spectrum analysis, and not a Quasi-Harmonic one, if that's what you're asking. (I do like to use a lot of noisy sounds though.) But ultimately, I'd like to know this info for sounds in the future. [This message has been edited by UmFuFu (edited 22 March 2007).] IP: Logged |
|
SSC Administrator |
posted 23 March 2007 08:15
In answer to your question, the FFT module does 1024 bands and the SpectralAnalysis model does 512. I asked about the nature of the signal in order to find out why you were looking for more bands of analysis. What is the fundamental frequency? Are there some artifacts you are trying to remove? (if so, there may be a different solution than simply adding more bands to the analysis) IP: Logged |
|
UmFuFu Member |
posted 23 March 2007 13:49
Hi Carla, thanks. (I think it's Carla!  )Anyhow, I find that the amount of FFT bands has a huge effect on the outcome of the resulting sound. I therefore find it important (as an artistic choice) to experiment with using different amount of bands to arrive at different esthetic results. IP: Logged |
|
cebec Member |
posted 09 December 2011 10:52
Sorry to dredge up this old post but I'm finding that a lot of my resyntheses sound wormy in comparison to those produced by other tools such as Csound's pvoc-ex analyses and pvoc syntheses, or using the SpectrumWorx plug-ins, for instance. I realize there are differences between the implementation Kyma uses and what these other tools are using but I'm wondering if there are ways around the current 512 band limitation (band-splitting?, roll your own with the FFT modules?) or if there are plans to increase the maximum number of bands available for the Spectral and FFT Sounds. I also realize there is a trade-off between frequency and time resolution and that the analysis/resynthesis is highly program-dependent but I'm finding that mimicking Kyma's settings as closely as possible in these other tools but increasing the bands to 1024, for example, has a dramatic effect on reducing the worminess. It is particularly evident when performing time-stretching. This is not meant to take away from these Sounds' overall ease of use and quality in other regards but I do find the limit of 512 bands to be somewhat frustrating. NB I am using a Capybara so I'm unsure if improvements to Spectral Analysis and their files in X.82 for the Paca/rana go some way toward addressing this. Thanks. IP: Logged |
|
SSC Administrator |
posted 09 December 2011 11:20
Some ways to minimize artifacts: Select source material with minimum of background noise or effects-processing. Select the spectral analysis (rather than the quasi harmonic analysis). Select Best frequency The Paca(rana) analyses have an improved signal to noise ratio over those that can be done with Capybara Select source material that is more pitched than noisy. You can probably pick up a few additional tips by watching Pete Johnston's KISS2011 presentation: IP: Logged |
|
cebec Member |
posted 09 December 2011 11:34
Thanks! I'd been looking for some tips such as those. A few quick questions: 1) To confirm: SpectralAnalysis Sounds are currently the best approach for complex or so-called program material over others such as Tau, GA, harmonic? 2) Are there any offline processes I can use to improve the analysis, such as amplitude normalization, resampling to a higher SR, noise reduction? 3) And last, I have been curious as to the meaning of 'improved SNR' for the Paca/rana analyses. I'm familiar with some of the technical jargon -- what is the noise, exactly, that was reduced and how does it affect the resulting resynthesis? Because, typically, I see reference to the time or frequency resolution, amplitude of side lobes, etc., when reading various papers about all of the different methods, so I'm curious what 'improved SNR' means in this context. Thanks, again! IP: Logged |
|
SSC Administrator |
posted 09 December 2011 13:11
quote: In general, the quasi harmonic analysis, Tau, and GA are designed for harmonic material. That does not mean it's not worth trying these on noisy material because sometimes it works. But yes, theoretically, non-pitched, non=harmonic sources are best analyzed with the spectral analysis Tool. Or LiveSpectralAnalysis (without the Harmonic box checked).
quote: Amplitude normalization might help. Up-sampling the original would not improve the analysis.
quote: Greater dynamic range and smaller quantization steps between amplitude values. IP: Logged |
|
cebec Member |
posted 09 December 2011 13:46
Great, thanks, and Pete's presentation was very interesting, indeed. IP: Logged |
All times are CT (US) | next newest topic | next oldest topic |
  |
|
Contact Us | Symbolic Sound Home
This forum is provided solely for the support and edification of the customers of Symbolic Sound Corporation.
Ultimate Bulletin Board 5.45c