What is a Sound Architect?

I coined the term Sound Architect to describe the work I do. The "Architectural" component of the term relates to my approach to sound, where I work from the ground up, so to speak, to develop low level manipulations of sound's primary elements. Essentially, I deconstruct, analyze and separate sound by recognizing individual events, elements, or spectral properties, and depending on the situation use the resulting components to modify existing sounds or reconstruct new ones. For example, I might separate a tone into its harmonic or partials or percussive components, and then rebuild those elements into something new.

In the early 90s I created the technology which enabled me to extract an extremely accurate MIDI or DNA Groove Template file from an acoustic performance. These DNA extractions, along with other forms of deconstruction, have allowed me remaster and remix from subtle to extreme manipulations, as well as create custom drone tones, and virtual instruments in the form of sampler patches. To accomplish this type of work a series of proprietary software processes (Mac OS) have been developed and these work entirely in non-realtime mode. These custom processes are not compromised by the limitations imposed by the realtime mode and often take 100-200 times longer than the length of the original sound in order to create the final sound.

Another important component of analyzing a direct sound is understanding its dynamic relation to enclosures. The results allow me to create extremely accurate (non-realtime) re-creations of the actual reverberation of many unique and aesthetically inspiring places, so a recording can be placed virtually into an acoustic setting of your choice. Some of the enclosures I have worked on include the King's Chamber inside the Great Pyramid at Giza, The ancient Pantheon in Rome, Giotto's Bell Tower in Florence, Palladio's San Giorgio Maggiore in Venice, the Crypt of the Pantheon in Paris, the Villa of the Mysteries in Pompeii and the Glen Gould Studio in Toronto, Canada.

As well as interior enclosures, I am interested in the dynamics of natural settings and exterior reverberation locations which I have also recorded and analyzed. Some of these include; The Valley of the King's at Luxor, the Streets of Venice, Ouimet Canyon in Thunderbay, a 4 foot crack in a massive rock formation located along the Northern Shores of Lake Superior, as well as several forests in Canada.

What I am particularly interested in now is applying aspects of the reverberent signature from sonically moving and visually beautiful spaces to musical passages in order to evoke harmonically "shadows" of these spaces in unpredictable ways.

How do you remaster using Sound Architecture?

Below is a description of a typical process and philosophy I employ when remastering. It pertains specifically to a CD remastering project I undertook for the CD "Sondes" by Electro-Acoustic Composer Randall Smith.

In remastering the works on Sondes I was primarily concerned with enhancing the clarity while revealing as much of the sonic detail as possible in each composition. To accomplish this I created a series of proprietary software processes to be performed in non-realtime mode. These custom processes are not compromised by the limitations imposed by realtime mode, and result in superior enhancements of sound quality. The tasks of each of these processes and their affects on sound quality and compositional detail are described below.

Equalization is the first signal processing performed. It was performed in 32/64-bit floating point with FIR filters (phase linear filters). Before optimizing the sound of each recording a very accurate analysis of what the actual overall spectrum of each track was first required. The strategy of the equalization was to balance all the frequencies present in the signal so that all the components would be heard with no dominate frequency band masking or overpowering any other frequency bands.

The removal of the frequencies below the audible range was also performed with high-quality digital high-pass filters. This was useful because a common result of intensive sound editing and mixing of music, especially in the composition of electroacoustic music, is the creation of sidebands, particularly in the frequency range between 1 and 20 Hz. Virtually any fade in/outs contain a broad range of frequency components with a pronounced emphasis in the sub-audio band. The manipulation of these 'control signals' (fade in/out) upon the audio signal superimposes the control signal spectrum onto the audio spectrum. The complete removal of these frequencies added clarity and dynamic range while preserving the loudspeakers when this music is played at high levels.

Reverberation was also added to the mixed electroacoustic works Continental Riff, Convergence, and Liquid Fragments, in order to add spaciousness and depth to there overall sound. A custom reverb signal, or impulse, was generated with spectral properties designed to enhance but not interfere with the direct (original) signal. The reverberation signals were mixed at low levels (-19dB) relative to the original signal. The reverberation signals have midrange and treble emphasis (with much less energy in the bass region). This type of reverberation spectrum was chosen to enhance the clarity of the transients yet add a spacious component to the overall sound. The stereo field of the reverberation signals was designed to minimize the image smearing by having the reverberation signals positioned at the same point in the stereo field as the source instrument. In other words, if an instrument sound was heard at the extreme left channel the reverberation of that signal was also heard only at the extreme left channel. The process of preserving the stereo field added depth to the composition while maintaining the clarity of the original performance and composition.

A dithering signal was used to convert the 3-bit floating point signal into the final 16-bit soundfiles for the final CD. A proprietary process of extracting the 8~12dB below the 16-bit signals noise floor (in the original 32-bit floating point soundfile) and adding this music into the lowest 8~12dB region of the final 16-bit signal was also performed.

Finally, the bringing together of these processes in an effective way was as much a pleasure as a challenge. The repeated listening required for this remastering gave me the chance to appreciate the rich sonic layers developed in Randall Smith's music. Working with contemporary composers gives me the first hand opportunity to witness as well as participate in developing technology for music.

How are Reverb Impulses Created?

The sample length of each reverberation impulse (RI) varies - the shortest Classical Hall at 44.1khz/16bit is 485kb; the 192khz/24 bit impulse is 6 mb. Most of the 176khz/24 bit reverb impulses are typically between 8-12 mb. The 44.1khz/16 bit RI's fall in the 600K to 2 mb range.

Considerable amounts of time, design, testing, energy, skill and capital were elemental in the development of this product. Here is a brief outline of theprocess - but please keep in mind that most of these stages include proprietary technologies that I have continuously evolved since the mid90's. First, the reverb impulses are not "starter's pistol" samples recorded in a hall. This and several other methods have been used to generate an acoustic recording of the RI of the space in question.

In the next step after the recording process, each RI is analyzed with custom software designed to reveal the "key ambient components" of the hall. Often, several RI's of the same acoustic space need to be analyzed before the "correct acoustic signature" is revealed. Intense processing time can berequired in this phase for just one "true" acoustic RI signature. The final stage reconstructs the RI in 64 bit floating point architecture with conversion into the SDII and Wav 16/24 formats. (Please note that the same RI at different sampling rates has to be recompiled from scratch and itswaveform optimized for the best sound at the given sampling rate.) If you listen to the same RI at 44khz, 48khz, 88.2khz, 96khz, 176khz and 192khz, all will sound similar but each will have its own unique and subtle sonic idiosyncrasies.

Processing/rendering was all performed on a 867MHz dual processor Power Mac G4 using OS X and System 9.22. This machine is about 4 times faster analyzing and reconstructing RI than my PowerMac 9600 with a G3 300MHz accelerator card. The signal processing for this and all my other sampling CD's since 1993 has been custom proprietary scripts developed using DSP Designer which works inside Apple's MPW environment . One round of calculations (110 RI's) took a number of "round-the-clock" weeks, at which point all data was recalculated again 3 additional times to derive the best possible, most accurate sounding RI.

The production of these Reverberation Impulses are the culmination of years of research, and as anyone who is involved in the field of research knows, return rarely equals the time invested.

How do Pure Space CDs Compare to other Reverb Impulses?

The principle of convolution offers astounding potential in the area of reverb impulses (RI) previously unattainable with existing reverb technology. Most (if not all) RI renderings of "high end" digital reverb widely available on the internet barely scratch the surface of what is attainable using the power and sophistication true 24 bit convolution can deliver. When using these "freeware" RI's, you are basically using yesterday's 16 bit technology on today's 24 bit system. It is not surprising that these "hardware simulations" do not have the neutrality/clarity or maximum richness of impulses that can be explicitly designed using convolution. Numerical Sound's impulses are the only impulses that have been optimized for true, full 24 bit performance which the following graphs will illustrate.

Six reverb impulses were analyzed and plotted for both frequency and dynamic range. Get the Graphs (opens in another window).

Results are provided in the following three graphic representations:

1. The first graph shows the decay characteristics in decibels of the reverb impulse. (Data is from the left channel). The red line in each graph is the decay rate a realistic acoustic enclosure should follow (down to the noise floor -144db for 24 bits).

2. The second graph illustrates the full frequency response of the reverb impulse. (Note: The frequency range of primary importance is below 1000Hz).

3. The third graphic zooms in on the high frequency response (from 1khz to 20khz). Reverb impulses, by their very nature, should have a jagged frequency response with a multitude of peaks and valleys, however the overall trend line (depicting an average of all these small variations) should be flat.

Given the state of today's technology, there is simply no excuse for a "non-flat" frequency response in a reverb impulse. It makes more sense for a producer to use an EQ plugin (or FFT filter impulse) to alter the sound of the wet signal and have it change dynamically with the musical nature of the mix than resort to various high frequency enhancers or multi-band dynamics processors in order to fix the overall sound.

An acoustic reverb impulse should decay at a constant even rate. When one looks at the first graph for each impulse the line should descend straight down to the noise floor. Impulses should look like the Numerical Sound impulse (#1).

The noise floor and/or smallest signal level in decibels (y axis vertical) versus the number of bits in the sound file (x axis horizontal).

What is Convolution?

Convolution requires two signals. The audio and the impulse signal. The convolution process take these two signals and produces a third new audio signal. The process requires a great deal of calculation. The length of time depends on how long the impulse and audio signal are.

The impulse can be almost any type of audio signal. The key thing to remember is that the impulse and the audio signal must have something in common to hear a result. For example if the audio signal has only a sine tone at middle C and the impulse has a sine tone at C# the result is no signal - just the noise floor of the audio signal

The impulse can be any type of signal but there are a couple of classifications.

Filter (not moving static)
Echoes
Reverb Impulse
Timbral filter
The filter impulse can be any conceivable type of fixed equalization (not a moving analog synth filter). These impulses sound like ticks and in effect have all the audio frequencies from 0-22KHz (44.1 Sampling Rate) so they always have something in common with the audio signal. They have everything in common the only difference is that the frequencies (or harmonics) of the impulse are all not at the same amplitude so the convolution process boost or reduces the gain which depends entirely on the make up of the impulse.

A tick (a single amplitude sample) in a sea of zeros before and after this tick will result in an echo of the entire audio signal. An impulse with such a tick at sample 1 then at sample 44100 but -6 db down and then another sample at 88200 but -12 db down will result in an an audio signal with the original sound and two echoes 1 second apart. The original signal then the original signal delayed 1 second but mixed in at -6 db original then another original signal delayed 2 seconds but mixed in at -12 db. This type of echo effect is heard in some manner in many digital effects units.

The Reverb Impulse class of impulses are noise like signals that decay over time. The advantage of a noise like signal is that is has all the harmonics in the audio spectrum. So regardless of whatever the audio signal has in it harmonic make up it it will always have something in common with the impulse signal. If the audio signal has a harmonic at middle C the reverb impulse also has that harmonic but it will last much longer than the audio signal tone (should be wetter sounding !) then the results will be a middle C tone that slowly decays over time (reverberates). The length of the delay at middle C depends on the characteristics of the reverb impulse.

The timbral filter impulses are very complex impulses designed to transform the timbre of the source audio material. There are no examples that I am aware of these type of tones on the internet. They take a great deal of time to create. What I personally do is sell the results not the impulses. There are result examples posted on the internet are samples from Drone Archeology, the Percussion Wall (some of the loops in 02pwall.mp3 <-- NOTE: needs to be renamed ---) and the Electro-Acoustic Modelling Examples of New Orleans; The Delta Grooves.

What is the difference between DNA Groove Templates and a Midi File that accompanies several drum loop sampling CDs?

Midi files are a representation of the arrangement of an acoustic performance of the drummer, DNA Groove Templates however are like a metronome in the sense that all the 16th or 8th note triple notes are uniquely defined in a time line/grid that matches the drummers feel.

Obviously drummers do not play all the notes in a bar (a good thing!) so match quantizing (transferring the timing) from one midi drum track to another musical part such as the midi base track will not work well because all the events are not defined in the midi part - this is where one should use the DNA Groove Template instead.