I have some room impulse response at 48kHz sampling rate. I have recording at 16kHz sampling rate.

I want to apply impulse response on the recording. What I do now is downsample impulse response 48kHz -> 16kHz and then convolve both signals.

Is it ok to do so? What are possible downsides of such approach?

PS. I guess the question could be restated as "is it ok to resample FIR filter coefficients?"


"is it ok to resample FIR filter coefficients?"


What are possible downsides of such approach?

Down sampling always contains trade offs: you will need to apply a proper low pass filter to deal with aliasing. The choice of low pass filter will determine that trade off between band, residual aliasing, time domain distortion, preservation of causality, transient behavior, etc.

  • $\begingroup$ Using an FIR filter with many taps to resample an impulse response would be a bad idea. The FIR impulse response would smear the room impulse response. $\endgroup$
    – Ben
    Sep 22 at 13:18
  • 1
    $\begingroup$ I wonder if it would make sense to first look at the impulse response's spectrum, and if it really doesn't have significant content above 8kHz, just sample it. The Nyquist criterion, after all, isn't about the filter you put in front of the sampler, it's about the frequency content of the signal at the sampler. (Note that I have no clue of the typical spectral response of rooms, so I could be suggesting something totally off base). $\endgroup$
    – TimWescott
    Sep 22 at 15:11
  • $\begingroup$ @Ben Room impulses responses are typically many thousands of samples long, so the smearing of the resampled is mostly harmless. Causality can be more of an issue there. $\endgroup$
    – Hilmar
    Sep 22 at 16:42
  • $\begingroup$ @TimWescott: Room impulse responses typically have plenty of content above 8 kHz. The only physical mechanisms that create a bit of a roll off are air absorption and most surface materials get a bit more absorptive. However, in residential rooms, that's only a minor effect. $\endgroup$
    – Hilmar
    Sep 22 at 16:45

Generally yes. Think about it in frequency domain: there, convolution is simply multiplication, and if either the signal or the impulse response have no energy above some frequency, the resulting output will not have any either.

The only issue is that impulse responses tend to be short, and resampling may use windowing and stretch them in time domain. Make sure to use padding as appropriate for your resampling algorithm and cut the result as needed. (Compare your original and resampled impulse response in time domain.)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.