I have a problem with the Renoise cabinet simulator.
I have my interface and Renoise set to 44.1kHz. If I set or render to 96kHz, I’m finding that the cabsim plugin is sounding disproportionately louder.
Researching on the forum here I found that the “impulse response” for the cabsim was originally sampled at 44.1kHz (not really sure what that means), so rendering or playback at 96kHz results in certain recalculations. I presume that it is these recalculations that are causing it to sound much louder.
Is there anything that can be done about this? I originally mixed my track at 44.1kHz, and I’m just wanting to export my track to 24-bit/96kHz for mastering (incidentally the track is entirely softsynths with no samples, so I would have thought this kind of sample rate change wouldn’t be a problem). I’d rather not have to remix it just to balance the cabsims’ volume for this 96kHz render.
Any help or advice anyone can provide would be appreciated! Thank you!
System:
Mac Mini 2.26GHz Intel Core 2 Duo
4GB RAM
Edirol FA-66 Audio Interface
Mac OS X 10.7.5
Renoise v2.8.1
AFAIK this device convolves input signal with chosen impulse response (convolution is basicly a sum of input_signal(sample)*impulse_response(sample) products). Both signals have to be of the same sampling frequency so each sample of an input signal can be multiplied by corresponding sample of an impulse response. If they are of different sampling rates (which is true if you change sampling rate of your audio interface) then the impulse response has to be upsampled or downsampled. It means it will contain the same/very similiar audio information (spectrum) while having different number of samples, a number that matches current sampling rate.
Thanks for this explanation. Does this mean that changing the sampling rate and having the cabsim get disproportionately louder is expected behaviour for this effect?
No. It means you can expect it to sound different though. But don’t expect this to always be louder, or similarly changed in tone!
A request has previously been made for the IR responses to be included for the common sample rates provided by Renoise (or at least the basic multiples, so interpolation should be quite basic and not sound quite as different.) You could try and render at 88.2kHz and see if it sounds more like you expect (being double 44.1 you would hope it would be more similar.) But really you should mix at the rate you are going to export at, because Renoise native effects are not the only ones which may suffer from this! (I think the Cabinet Simulator is the only one of the native effects that does…)
I think it may be some bug in up/down sampling functions. Higher amplitude means that probably normalisation went wrong at some stage. But its just a guess.
Yes, actually I mixed at 44.1kHz because that was the rate optimal for the music’s primary purpose; it’s game music, and 44.1kHz is optimal. The reason I wanted to export at 96kHz was because we’re going to be making the tracks available for purchase separately via various download vendors, and for that I’m going to be sending the tracks off to a mastering house. As the music is entirely composed of soft-synths (i.e. no samples), I figured that I could render the tracks out at 96kHz to send off to the mastering engineer. I didn’t realise that changing the sampling rate would affect plugins so drastically. Lesson learnt!
So is there still the possibility that this is actually a rather perverse bug in my beloved Renoise?
If the only difference is the Cab Sim then you can cross your fingers that the IR curves for different rates are included in the forthcoming version but I wouldn’t hold your breath…
Bear in mind that soft synths, such as every instrument you have just stated you use in all your songs, may also sound different at different rates! Are you positive none of these have been affected by the change?
I’m positive only to the degree that everything else subjectively sounds the same. Although as you point out, it’s highly likely that there are other small differences deep in the mix that I’m possibly not noticing.
I suppose in this instance I’ll just have to settle for sending the mastering engineer 44.1kHz files to work with, just to be totally confident he’ll be getting exactly what I tracked and mixed. That’s no huge issue. I started this thread simply wondering if there was anything I could do to easily remedy the situation.
Hopefully we may see a fix for this at some stage! Until then, thank you all for your help.
I guess they want 96kHz soundfile for better precision of their mastering software. In this case they can just upsample your 44.1 kHz material without quality loss to 88.2 kHz or with little (probably not noticeable) difference to 96 kHz.
Yes it is possible without a loss of quality, but you won’t get an improvement of quality either. You can put 1 liter of water into 2 liter bottle without losing any water, but you won’t gain any more water by doing it either.
This indeed, but that does depend on whether the soft-synth won’t support more than 44Khz. If you have softsynths that do support 96Khz, they might offer a higher integrity levels for your sounds, but theyn you also have to mix your songs in 96Khz as Kazakore already advised.
The real question is: is it possible to design an ideal digital filter? Of course not.
It works like this:
If you want to downsample by factor of 2, for example from 44100 to 22050, you have to take a signal with sampling frequency equal to 44100 and take it’s first sample, the third sample, the fifth, the seventh… etc. You end up having a signal which has half as much samples as the original. Before you do that you need to be sure that signal has no frequency above new Nyquist frequency (which is half of the sample rate). If it does, you have to apply a lowpass filter to remove it. Otherwise aliasing will occur. So by downsampling you loose all information in the spectra which is above 10025 Hz (half of the new sample rate) and also some of the higher frequencies below 10025 which will be distorted because the filter is not an ideal one and it has a slope in its amplitude characteristic and phase shift in its phase characteristic (although some filters have linear phase for example Finite Impulse Response filters can be designed to have linear phase). The process is sometimes called decimation.
If you want to upsample by factor of, let’s say 3, for example from 10025 to 30075, you have to insert three zeros between every sample of the original signal. As a result you will obtain a signal of sampling frequency 30075. But if you look at its spectrum you will see it changed. Additional information is present - a copies of the original spectrum. You have to remove them using lowpass filter. When you do this these copies will vanish. After removal of those spectrum copies you will see that zeros you inserted in time domain changed into other values in a way that signal is smooth. The process is thus called interpolation because you interpolate a new, smooth signal, by removing spectral copies with lowpass filter. What information do you lose? As before, lowpass filter has some slope and phase shift - it will change signal a bit in its higher frequencies.
If you want to upsample signal by a factor of 3/2 you have to upsample it by factor of 3 and then downsample it by factor of 2.
So, when you want to change frequency from 44.1 kHz to 96 kHz - you have to upsample it by factor N, then downsample it by factor M. You are filtering it with LP filter twice. N and M are such constants that N/M = 96/44.1 = 2.17. The closest I got was 11/5 = 2.2. But this is just a guess
If you are interested in this topic I highly reccomend this book: Amazon.com
