Hello;
This is a very interesting tool; already got good results with some of my mixes revisited.
Just one question:
If I save a processed file (name_eq.wav) and load it as a reference again the curves should match each other.
When I do so I see some differences in the lower frequency range. Is this due to the resolution of the FFT algorithm (I think this must be around 5Hz at 44100 samplerate and 8192 points) or am I doing something wrong here?
Another point is the procedure of saving files:
In "file -> EQ output" I have the option to save but nothing except the file format settings seem to be saved here.
The "real" saving is with the record button - thats a bit confusing to me (please don't take it as criticism - just a thought).
Regards
Ulrich Klauer
About FFt resolution and saving files
-
mflorio
File-EQ-Save
I agree. The EQ - "Save" is misleading. Perhaps popping this dialog when the record button is pressed makes more sense.
-
Har/Bal
About FFt resolution and saving files / File-EQ-Save
Yes, your thinking is correct.
The low frequency discrepancy between what you chose the spectrum to be and what you actually end up with is down to the filter being an 8192pt FIR. For 44.1kHz sampling the resolution is around 5Hz so if your filter design has significant variation within a bandwidth of around 5Hz the filter design will not match it well. This is where the descrepancy comes from.
The only solution to this (when using an FIR realisation) is to increase the length of FIR, which also increases latency. We took the view that 8192pt was about optimum in terms of latency and resolution and our limited listening tests demonstrated little added benefit from have a better resolution than 5Hz.
The EQ Output/Save is confusing, I agree. It stems from using a common dialog box to implement it. The Save button needs to be renamed to OK to remove the confusion.
Regards,
Paavo.
The low frequency discrepancy between what you chose the spectrum to be and what you actually end up with is down to the filter being an 8192pt FIR. For 44.1kHz sampling the resolution is around 5Hz so if your filter design has significant variation within a bandwidth of around 5Hz the filter design will not match it well. This is where the descrepancy comes from.
The only solution to this (when using an FIR realisation) is to increase the length of FIR, which also increases latency. We took the view that 8192pt was about optimum in terms of latency and resolution and our limited listening tests demonstrated little added benefit from have a better resolution than 5Hz.
The EQ Output/Save is confusing, I agree. It stems from using a common dialog box to implement it. The Save button needs to be renamed to OK to remove the confusion.
Regards,
Paavo.
-
ukm
About FFt resolution and saving files / File-EQ-Save
Hello;
thanks for quick reply - that makes it clearer.
Great tool !
Regards
ukm
thanks for quick reply - that makes it clearer.
Great tool !
Regards
ukm
-
Anticipator
Re: About FFt resolution and saving files
"...5Hz so if your filter design has significant variation within a bandwidth of around 5Hz the filter design will not match it well..."
what does that mean for practice?
How can I design a filter that will match?
what does that mean for practice?
How can I design a filter that will match?
-
Har/Bal
Re: About FFt resolution and saving files
To give a concrete example, if you had a recording that had a 50Hz/60Hz constant tone (ie. hum) the true spectrum would correspond to a very narrow line around 50Hz (true because Har-Bal actually displays a smoothed spectrum to make spectrum adjustment easier - ie. too much information otherwise).
Now if you were to design a filter with Har-Bal to attenuate this spectral spike you could but any frequencies within a bandwidth +/- 2.5-5Hz from 50/60Hz would also be significantly attenuated. That is the effect of this limitation. The only way to reduce this is to increase the filter size from 8192pt to something bigger.
In practice there isn't much to be gained by this because for almost all audible frequencies this level of selectivity is better than the critical bandwidth of human hearing. The net effect of this is that a longer FIR filter will perform better in frequency matching but it is unlikely that you will hear a difference in the performance, or at least not a significant one.
I hope this helps.
Regards,
Paavo.
Now if you were to design a filter with Har-Bal to attenuate this spectral spike you could but any frequencies within a bandwidth +/- 2.5-5Hz from 50/60Hz would also be significantly attenuated. That is the effect of this limitation. The only way to reduce this is to increase the filter size from 8192pt to something bigger.
In practice there isn't much to be gained by this because for almost all audible frequencies this level of selectivity is better than the critical bandwidth of human hearing. The net effect of this is that a longer FIR filter will perform better in frequency matching but it is unlikely that you will hear a difference in the performance, or at least not a significant one.
I hope this helps.
Regards,
Paavo.