Comparing a T600 impulse response with a free version of the same preset
By Peter Emanuel Roos, producer of the Samplicity IR libraries
This is a revision of an article that I posted on the Samplicity site when I launched the first Samplicity library Vol. 1 T600 Impulse Responses. It is meant to document some “lessons learned” before and during the project and certainly not to bash the makers of other impulses responses, that have also invested time and energy in the products that they share for free on the web.
“I was struck by how disappointing so many free IRs are”I have been asked several times what the differences are between my impulse files and corresponding files that can be found in free but smaller downloadable archives, in particular from www.noisevault.com.
When I first started to compare free IR files with an excellent commercial library that I have from Ernest Cholakis, I was struck by how disappointing so many free IRs are. They often have a very specific “color”. They can have have high noise levels and tend to be simply truncated at the end, instead of being nicely faded out just before the background noise takes over.
So when I started to think about creating an impulse library for myself, I wanted to go for maximum quality, resolution, lowest noise levels and for a pristine and transparent sound that would ressemble the original as much as possible.
For the hardware links I used two separate DAW’s with RME audio cards. They were connected via ADAT to an RME ADI-8 DD digital format converter, which in turn was linked with AES/EBU cables to the digital reverb. All devices were synced with RME Word Clock. For playback and recording I used SoundForge 6 on both DAWs. Audio processing was done in 32 bit floating point format, using Waves plugins in SoundForge. I decided not to use my Logic or Cubase SX multi-track applications to keep the digital signal path as simple and clean as possible. 44.1 and 48 Khz versions were recorded and processed separately. No automatic batch processing was used, all sweep recordings and resulting impulse responses were processed individually, taking their characteristics into account and making sure that their variations (left, center, right) were processed in exactly the same way.
Using white noise for calibrations
During the recording sessions for the T600 library (already back in 2005),
“I already knew that I was going to need drastic EQ corrections to handle deconvolution side-effects”I also recorded mono white noise bursts through all the presets for calibration purposes, along with several other source sounds and timing signals. I wanted to have these “all wet” noise recordings to help me compare the frequency spectrums of the original device’s programs and my final IRs. From the literature I already knew that I was going to need drastic EQ corrections to handle deconvolution side-effects.
With these dry and wet noise files, it was now also easy to compare my final IRs with other, free IRs from the same digital reverb. The results confirmed my earlier disappointments with free IRs: they can be very colored indeed, they can have rather high noise levels (around -50 to -70 dB SNR) and they do not have gentle fade-outs.
For these comparisons, I used the Church program and ran the original, dry white noise through the original hardware reverb, through my T600 IR version and through a free IR found on NoiseVault. I have captured a few screenshots from the frequency plots.
“The free IR version however has a serious bump in the middle frequencies, corresponding to its boxy sound” The graph above shows that the Samplicity IR has a very similar spectrum compared with the original device preset, in the graph right above this one. In the upper range, the Samplicity IR seems to be slightly brighter (in a range that most cannot hear). The drop below 30 Hz is the result of a very steep low cut to remove deconvolution side-effects, which introduced a lot of subsonic energy in the IR’s. After removing them, the signal-to-noise ratio very significantly improved.
The free IR version however has a serious bump in the middle frequencies, corresponding to its boxy sound. It also has more high frequencies than the original Church program, making it sound much too bright for a real church. Also note the strange peaks and valleys in the spectrum. This is often a result of using the “spiking” instead of the “sweeping” method.
“To spike or sweep, that’s the question”
On a lot of forums you can find posts saying that all you need to record impulses responses is to run a Dirac spike (a single sample of maximum volume) through a digital reverb unit to capture its sound. From a mathematical perspective this approach may seem plausible, but it completely discards the serious side-effects introduced by the poor energy in a single sample. Also, even digital hardware devices are simply not designed to handle such extremely short spikes. The sweeping method is the only way to obtain good results. Its secret lies in the time domain: a single spike can be as short at 1/48000 of a second… Imagine how little energy and information is carried in a single spike compared to a sine sweep of 15 to 30 seconds!
The reason why sine sweeps are hardly used for free impulse responses is simply that it takes so much time to record, edit and deconvolute them! Another reason is that it requires extremely careful synchronization between the sweep signal and the recording in order to “deconvolute” them. For the Samplicity IRs an approach has been devised that guarantees sample-accurate synchronization. This is one of the reasons why the Samplicity impulse responses sound so extremely close to the original hardware reverb units.
I also zoomed in on the start of the IR files in SoundForge. From the wave form is obvious that there already is high frequency content (noise) present around the very first reflections in the free IR, whereas the T600 version displays very distinct and clean spikes, representing the first reflections (echoes) of the original signal. This noise in the free IR is actually starting before the first reflection peak, supporting my suggestion that it is noise what we see here. Furthermore, the first reflections in the IR are not at their correct position: the leading “silence” has been removed. This missing leading silence is actually the “pre-delay” as configured in the preset. So, it really should be present in the IR as well. I discovered that this actually holds for most of the free IR versions of this digital reverb. In the T600 library the embedded predelays, or silences to the first reflections, are accurate to a single sample with respect to the original program.