Strange Convo is a Max4Live audio effect created by Tyler Sadlier that uses pseudo-realtime convolution with a sampling paradigm to dynamically filter an audio signal. It allows an Ableton Live user to sample stereo audio into the device’s buffer which is then convolved with incoming audio. The convolved audio sounds similar to a resonator, as the common frequencies between the sample and the real-time input are amplified and the uncommon frequencies are attenuated. Unlike a resonator, however, the resonant frequencies change with the buffer sample’s timbral qualities through time. Strange Convo is particularly good at smearing two harmonic sources into ambient washes, or applying tonal/chordal qualities to percussive sounds. By encouraging experimentation with different sample sources, Strange Convo pushes users towards creative new applications of convolution cross-synthesis.
Phase 1: Researching
I spent weeks 1-3 of the quarter researching potential project paths. I brainstormed several ideas, including an interactive synthesizer tutorial website made with WebAudio, an audio effect plugin that emulates the sound of a laptop mic run through Skype (a modulation, filter, and distortion multieffect), and some sort of unique application of convolution. With each of these potential routes, I looked into the surrounding literature and libraries online. For the WebAudio project, I spent several days experimenting with simple WebAudio features until I found a resource called Tone.js that makes synth and effects creation significantly easier/faster than just WebAudio itself. With my concurrent research into convolution techniques, I realized that I could easily record an impulse played back through Skype (and/or record using a laptop mic) to get a static snapshot of filtering artifacts (but not time-based irregularities). Capturing an IR through Skype could help me further uncover the artifacts' timbres and hopefully emulate the sound through other means.
In my further research of capturing impulse responses and convolution reverb techniques, I found several websites that recommended using non-impulse response sources in convolution reverbs for creative results. Most convolution reverbs utilize a file browser or a drag-and-drop paradigm for loading new impulses. It occurred to me that if a convolution effect did not prioritize using standard impulse responses, the method of loading or capturing the "impulse" could be completely different. It was at this point that I knew I would create a convolution effect with an interface built around sampling. Furthermore, the idea of convolving two real-time audio signals seemed like it would produce incredibly unusual results. I sought to research different applications of audio convolution that could help me with creating my effect.
I decided early on that I wanted to use Max4Live to build this effect. Luckily, I found a set of Max externals developed for a variety of convolution related tasks distributed under a modified BSD license: the HISSTools Impulse Response Toolbox. After inspecting all of their externals in Max, I found that the multiconvolve~ object was most relevant to my design. This object combines "time domain convolution for the early portion of an IR with more efficient FFT-based partitioned convolution for the latter parts of the IR." Because of the multiconvolve~ object's design, it cannot convolve two real-time signals and needs the "IR" to be loaded into a Max buffer~ object. With this in mind, I began focusing exclusively on designing the sampling paradigm.
Phase 2: Designing and Prototyping
Phase 3: Building and Tuning
Reflections / Future Plans
One of my initial visions for Strange Convo was to convolve two continuous signals in real-time. However, the design of multiconvolve~ and the limitations of Ableton’s audio routing in Max4Live complicate this goal. HISSTools’s multiconvolve~ external object requires the source or "IR” sound to be stored in (a) buffer~ object(s). My “Continuous” button was a fast workaround for implementing pseudo-realtime convolution between two stereo signals, as it continually records into the buffer and replaces previous audio. Since I capped the buffer length for this mode at 500ms, the “source” is delayed by half a second. Adding an online buffer length control to the user interface could make this feature more expressive and useful for different forms of sampled sources.
Unfortunately, the utility of this Continuous feature is limited by the fact that Ableton does not easily allow audio to be routed from a separate track to an M4L device. Since Strange Convo's buffer~ can currently only sample from the same track that it resides on, the result is an audio track convolving with itself from 500ms ago. This feature certainly sounds interesting with some sources, but implementing a workaround for streaming audio between tracks would open up many possibilities. Granulator II by Monolake is a granular synthesis Max4Live instrument that can sample audio from other tracks in an Ableton project. To accomplish this, Granulator II uses an “Input” audio effect that stores audio into a global buffer that the Max4Live instrument can load and access. My biggest hope for a new iteration of Strange Convo is to add this multi-device sampling functionality (along with buffer control) to enable convolution with other sources in near real-time.
Some other parameters I hope to add to Strange Convo in the future are one-knob controls for Compression, High-cut, Low-cut, and Pre-delay of the wet convoluted audio. Gain and Width already provide a surprising amount of control for balancing the presence of the convoluted signal in the mix, but additional effects could make sculpting the sound significantly easier. Interesting offline controls could include an envelope control for the sampled audio, as the transient shape of the buffer significantly changes how the convolution reacts. However, this would necessitate reloading the buffer every time the parameter is changed (as well as copying new buffers from the original buffer). Other fun features I'd like to implement are an oscilloscope, trails mode (to prevent cut offs of old reverb when sampling new source), sample waveform display, grainstretch, and a gate/threshold at which audio is automatically grabbed.