Jump to content

Alexey Lukin

  • Posts

    18
  • Joined

  • Last visited

  • Country

    country-ZZ

Retained

  • Member Title
    Newbie

Recent Profile Visitors

The recent visitors block is disabled and is not being shown to other users.

  1. I think that it's unrelated. This paper describes closed-form notch filters that are completely different from oversampling filters.
  2. Honestly, from their description I don't see much benefit. “Closed form” means that you can describe the filter with a simple formula, as opposed to calculating it as a result of the optimization algorithm. Many competing SRC filters, including iZotope SRC (in some modes), are closed-form filters too. The fact that they retain the original samples and insert (interpolate) the new samples between them may sound significant. But I believe that it doesn't make much sense in the audio world. Again, some competing filters work like that and there are good ones and bad ones among them (just think of linear interpolation). What matters more is that their filter is “optimized in both time and frequency domains”. It means that the designers are conscious of ringing and frequency response. In competing filters, the trade-off between the time and frequency performance is set by the filter length, which affects the cutoff steepness and the amount of ringing. The fact that a certain filter is “optimized” in both domains simply means that the designers have chosen a certain balance between time and frequency performance. It does not automatically mean that the filter is better than what you can achieve with competing products, esp. when you have access to the steepness control.
  3. pj, the amount of spread may also be related to presence of signal onset/decay within your analysis selection. When making a selection, be sure to exclude the signal onset or decay from it. Otherwise they will produce a similar spectral pattern which can be confused with jitter.
  4. Honestly, no idea. I don't handle the business side of licensing, my part is just DSP algorithms.
  5. The reduced ringing benefit only applies to intermediate-phase filters, i.e. those whose Pre-ringing is between 0 and 1. Linear-phase and minimum-phase filters only have the other two improvements.
  6. It shouldn't be hard to do, if 3rd parties request iZotope.
  7. The updated iZotope SRC in RX 4 now has more gradations for filter Steepness, fully compensated latency (for easy null-testing), and less time-domain ringing in intermediate-phase modes.
  8. I don't personally handle licensing, but it shouldn't be hard if he contacts iZotope. Yes, it should be possible, provided that there's enough DSP power. Again, iZotope would be the right point of contact for such collaboration.
  9. In the next version of RX, you'll be able to type .5 numbers in the Steepness field. The latency will also be fully compensated to simplify the null-testing (no more fractional latency).
  10. Create a unit pulse by silencing a few seconds of audio, then mouse-dragging a single sample upwards. Then upsample this pulse by a factor of 10 or 20 (by typing in the required rate). That will give you the impulse response.
  11. No, it's right: the order is approximately 4 times the steepness. For FIR filters it's not about dB/oct, it's more about frequency resolution measured in Hz. The reason why you see ringing is the bandlimited interpolation in RX: it plots the expected output of the D/A converter. If you only want to see ringing present in the filter impulse response itself, temporarily change your Waveform interpolation order (found in Preferences > Display) to 1. I'm usually too lazy to upsample for casual listening, but when I need to upsample I typically go with settings close to defaults. I think that different SRC settings are a matter of personal preferences and I do not have strong preference for one or another setting. That's why iZotope SRC provides all the settings to tweak. Yes, you should be able to open files, analyze the spectrum, change SRC parameters, inspect the curves, and save screenshots in the demo mode. For this parameter the higher, the better. However it only matters with “tricky” combinations of sampling rates, like 44.1k to 192k. When the max filter length is not enough, the output signal may have periodic minor crackle (usually too low to be audible, like –100 dB relative to the signal level). However if you are upsampling 2x or 4x, this should never happen, unless you are using insanely high Steepness. So, for 2x or 4x this parameter should have no effect, which can be confirmed by a null test. These settings are pretty close to one another already, but we'll consider adding more flexibility for Steepness. Thanks everyone for the feedback!
  12. We used to have a floating-point steepness prior to 2005, but changed it to integer-only since then. Try a null test of a .5 value with nearest integer values. If there's a demand in finer control of steepness we can consider it for future updates.
  13. For “the best of both worlds” the settings are: Steepness = 3, Cutoff shift = 1, Pre-ringing = 0. Actually steepness = 3 is slightly too steep, but steepness = 2 is slightly too shallow. The Ayre's filter is in the middle between those two. Here is the comparison of Steepness = 3 vs. Steepness = 2: As a side note, Ayre's filter is not really apodizing. Apodizing filters fall to –100 dB at Nyquist, not –6 dB. Steepness controls the filter order (something like order = 4*steepness). However unlike IIR filters, FIR filters used in SRC do not have a fixed dB/oct rating in the transition band. You can see that the frequency response curve bends and the slope gets steeper as the frequency rises. No, Pre-ringing and Cutoff Shift are floating-point numbers and are only truncated on a screen, not in the algorithm. Steepness is an integer.
  14. I'm typically using default settings: Auto-adjustable STFT and combined Spectrogram+Waveform view. In this mode, when you zoom in time, the time resolution of FFT automatically improves and you can see clicks more precisely. When you've narrowed down on a click, the waveform view will help you find the precise boundaries. There's no need to be super-precise with your time selection: the interpolation is usually quite good and will not distort a few extra selected samples.
  15. >> Although the graph only extends to -35 dB, you can move your cursor around the graph and the blue number at the top shows the attenuation at the frequency where your cursor is pointed. You can also scroll the scales by dragging or mouse-wheeling them.
×
×
  • Create New...