02. 2012. KORG MONOTRON. IMPULSE RESPONSE MEASUREMENTS ... state
that the Monotron uses the same analog filter as the 'legendary' Korg MS-10.
Karl Karlsson Institute of Sonology Signals and Systems II 21. 02. 2012
KO R G M O N O T R O N IMPULSE RESPONSE MEASUREMENTS
The Goal The goal of this assignment is to put into practice what we have learned this year in our Signals & Systems class by measuring the characteristics of a system by making an impulse response of it. In theory this impulse response can be used to simulate the effect the system has on a signal that goes through it by means of convolution. The system in question must be a causal and linear system for this theory to be true.
The System
I chose to measure the voltage-controlled filter of the Korg Monotron analog synthesizer. The Monotron is a tiny inexpensive synthesizer that most would consider more of a toy than a serious instrument but is has an undeniable charm both in the way it sounds and looks. The filter has an auxiliary input and can be used to filter an input signal independent of the rest of the synthesizer and is therefore ideal for measuring. Korg proudly state that the Monotron uses the same analog filter as the ‘legendary’ Korg MS-10 and MS-20. However the filter has a a definite ‘lo-fi’ quality and produces a lot of noise. For this reason it cannot be used to process sound in the same quality as the older Korg synthesizers but I
find the effect quite charming and interesting. I am hoping to capturing the imperfections and toy-like quality of this otherwise powerful filter. The filter is a low pass filter and I expect the outcome to be a low pass system with some irregularities.
The Process The measurements were made as follows. 1. An exponential sweep was generated using Praat. 2. The sweep was imported into Audacity and edited. 3. The sweep was played using my audio interface which was connected to the input of the Monotron. 4. The output of the Monotron was recorded through my audio interface using Audacity. 5. The recordings were trimmed and imported into Praat. 6. The recordings were convolved with a reversed version of the original sweep.
The Setup The Monotron filter has two adjustable settings that can be adjusted using knobs. These settings are the cutoff frequency and the peak (resonance). I decided to leave the resonance at it’s lowest setting and measure how the filter behaved with different values for the cutoff frequency. I made five recordings each with a different cutoff frequency. The knob that controls the frequency has no values printed next to it so I do not know the exact frequencies but divided the range of the knob visually. The settings are as follows. 1. 100% of maximum. The filter should not influence the sound. 2. 75% of maximum. 3. 50% of maximum. 4. 25% of maximum. 5. 0% of maximum. The sound should be completely filtered out. Using Audacity I recorded each setting using the sweep I generated with Praat as an input signal for the filter. The playback and recording was done using a Motu Traveler audio interface. I had to connect the Motu’s jack output and input to the Monotron’s mini-jack input and output using a converter.
Using Praat I time-reversed the original sweep and convolved it with each of the recordings I made. When convolving I set the amplitude scaling to ‘peak 0.99’ to normalize the otherwise way too loud result.
The Sweep The original sweep was generated using this Praat script available on Peter Pabon’s website. startFreq=10 endFreq=20000 lnRange=ln(endFreq/startFreq) periodT=15 multFact=lnRange/periodT Create Sound... ExpSweep 0 periodT 44100 exp(multFact*0.5*(xperiodT))*sin(2*pi*startFreq/multFact*exp(multFact*x))
This script generates an exponential sweep from 10Hz to 20kHz over the period of 15 seconds using a sine-wave. I slightly modified the original script to increase the range of the sweep since I was measuring a low-pass filter. The sweep is exponential so that the frequency resolution is perceptually balanced. For the measurement to work the amplitude spectrum must be flat. To ensure this a formula is included in the script so that when the frequency increases with a certain factor, the amplitude will increase with the square root of this factor. Otherwise the lower frequencies would have a much higher amplitude due to the time spent in those frequencies during the sweep. Using Audacity I created a short fade-in at the beginning of the sweep as well as a fade-out at the end of it. I did this to make sure there would not be an audible click at the beginning or the end that would affect the frequency content of the measurements. I also added 2 second of silence before and after the sweep for personal reasons.
Problems and errors As mentioned before the Monotron emits a constant noise. I made a conscious decision not to use any kind of noise reduction on the recordings as I thought that would affect the accuracy of the measurements. I also naively hoped that the noise would become part of the impulse response thus be reproduced when convolving it with another sound material. Instead the result was a downwards filter sweep of noise over the whole period of the impulse response (twice the length of the original sweep). This makes perfect sense as the noise consists of all possible frequencies and when convolved with the reversed sine-sweep would result in a filter sweep of those frequencies. When the impulse responses are convolved with different sound material the result sounds more like a cloud of sine waves than the original noise from the Monotron. A solution to this would be to record the sweep responses a number of times (the more the better) and then adding these recordings together. This would amplify the sweep signal more than the noise, as it is uncorrelated, and therefore leave a cleaner sweep response. The noise gets filtered along with the input signal although it seems to keep more of it’s amplitude. As a result of this the last measurement (the 0% cutoff) contains more low-frequency noise than the actual sweep. The impulse response does have an impulse in the middle, but when this impulse response is convolved with another signal the result is a wonderful, albeit scientifically useless, low-frequency mess.
Impulse Responses
IR 1 - 100% Cutoff
IR 2 - 75% Cutoff
IR 3 - 50% Cutoff
IR 4 - 25% Cutoff
IR 5 - 0% Cutoff When looking at these impulse responses the noise is very visible in the spectrogram. The amplitude is not very high except in IR 5 but is obviously there.
Results As I mentioned in the beginning I was interested in capturing the lo-fi characteristics of the Monotron while also capturing the ‘powerful’ filter. By convolving different sound material with the untrimmed impulse responses I have learned that the low-pass filter itself is pretty well simulated using this method except for the strange sounding noise that is created due to the noise in the recording. The filter however seems to be lo-fi enough as it is when the spectrum of the impulse responses are analyzed. For example the frequencies start to roll off at 3000Hz even when the cutoff is set to 100%. In this sense the filter behavior was very close to what I had expected. It had the obvious behavior of a low-pass filter but with some irregularities. When the impulse responses have been trimmed their effectiveness is greatly increased because the noise is decreased. When convolving sounds with these impulse responses the results are much cleaner. Impulse responses and convolution appear to be a rather effective way of capturing characteristics of a system and had I used the method I mentioned for reducing the noise I am sure that the system would have been captured fairly accurately. The noise if of course non-causal and therefore it should not be part of the measurement. The noise did however produce some interesting sounds in the end, especially in IR 5 where the cutoff was set to 0%. I think it might be interesting to look further into what is happening there and the possibility of using these methods for generating sound structures. I also want to mention that it was liberating to finally put these theories into practice.
