The data used in RISE: Sonic Sketches of Sourdough Cultures is depicted in the graph you see below. This is the Optical Density growth profile over a 48 hour period for the 8 most prevalent strains of yeasts and lactic acid bacteria (LAB) found in The Sourdough Project’s 500 starter samples. Using these data defined shapes was suggested by their similarity to the motifs of Terry Riley’s In C, a piece that continually shapes and sharpens my appreciation of timbre and harmonics.
This data set turned out to be less important in the great scheme of the final Sourdough Project paper, however by assigning chromatic pitches to the OD levels from the lowest measured amount (.0867) to the highest amount (.8816) among all 8 taxa, a unique motif emerges for each one. The intervals between sampling points/tones reveal the growth rate and expansiveness of each taxa. The notes at each sampling point when strung together create a pentatonic pattern spread out over four octaves that will be the sonic profile of each strain of yeast and LAB. Here is an example of the motif for L Sanfrancisensis, a lactic acid bacteria common to sourdough starters.
There were 40 density amounts over 4 octaves, so 10 notes were needed in each octave and two notes had to go. Leaving out C and F in a scale with G as the fundamental pushed the scale toward more dissonance, which helps to create the “sour” part of the sound. The chromatic scale runs from G0 to G5 (the scale runs from G0 to F#4, and then jumps to G5. G5 is heard only in W. Anomalus). Here is the piece that introduces the yeast voices and pattern profiles – String of Yeasts
The LAB voices are horn, synth, brass and a plucked resonant instrument. LAB do not reach levels higher than .5 on the OD scale thus are lower in pitch class range overall. Several of the yeasts soar into the 4th octave, but the LAB all stay in the 0-3 octaves as they grow slower and less abundantly.
And then there are the Acetic Acid Bacteria that have not received much attention in previous research. One of the findings of the Sourdough Project is that highly variable abundances of AAB are a key driver of functional diversity across the 500 starters in the study. The AAB also contribute heavily to starter aroma. In the soundscape AAB will take the form of sculpted noise- mixing various shades of noise with audio of watery bubbling sounds. And since AAB are drivers, percussion will be used as well. The primary AAB, Acetobactor Malorum, is represented by a polyrythmic frame drum statement.
The Yeast and LAB sonification profiles are what I call “data-driven” in that specific data points have been used to depict each Yeast and LAB voice. The AAB sonification is “data-derived” in that the use of percussion as a driver, of burbling, watery sounds as fermentation, and of post-soundscape frequency artifacts as VOCs were all suggested by descriptions of AAB in the published paper.
Three individual starters were sonified for the album. SD_522 was chosen because it may demonstrate the impact of Acetobacter Malorum on functional diversity in starter microbiomes. This starter had 6 of the 8 articulated taxa in measurable amounts and Acetobacter Malorum as the primary AAB. SD_131 contained Acetobacter Malorum and hit 4 of the 6 aromatic notes, so the last 30 seconds of the soundscape are the audio artifacts representing volatile organic compounds (VOC). SD_299 was chosen because it is mostly LAB and DOES NOT have any S Cerevisiae and very little AAB. This allowed me to play with a very different sonic pallette from SD_522 and _131.
The album is available March 30, 2021 on Bandcamp, and within the month on all other music platforms! Thank you for your support!
After reading and studying the data (so far) from The Sourdough Project, a bit of it jumped out as a possible sound pallete. The growth profiles of the five most prevalent yeasts and aabs (acetic acid bacteria) measured as increasing Optical Density over a 48 hour period. Measurements were taken in 12 hour increments and recorded from 0.1 to 1.2 levels of density.
I was drawn to this data because the graphs reminded me of waveforms.
I am not at liberty to reveal the details of the data, so suffice to say that these are 5 strains of yeast. We will call them pink, blue, orange, green and neon. The pinpoints mark the 12 hour samplings of the prevalence of the strain. So at 12 hours pink grew to around .25 OD, while neon grew to .6 OD. How to represent this in sound is the next question!
My old friend, the piano keyboard, provides a familiar sonic framework. A two octave chromatic scale will represent the sound of OD growth by stretching the OD measurement scale over the two octaves. Like this:
Each OD amount covers 2 notes. D and D# represent the .1 amount, E and F are .2 and so on. This allows some wriggle room when the 12 hour sample seems to be between two numbers as is seen with pink. The growth range for pink will run from D to F and encompass 4 notes. In the case of neon, the growth range runs from D to C and encompasses 11 notes. The differences in the growth rates will be heard in the number of and duration of the steps taken within each twelve hour time frame. So far, so good!
The time frame will run in beats and measures. Since it is 48 hours of growth, one hour can equal one measure. The step patterns will run up to the highest note indicated by the OD data at that particular 12 hour marker. That makes each sampling unit 12 measures in length – seems perfect. Even better, at 4/4 time, each 12 measure sampling unit is 48 beats long! Synchronous!
Lets lay out the first 12 hours of pink and neon. Since all the yeast densities begin from .1, all the patterns will begin with D in the 3rd octave (D3). pink grows from D through D#, E, and lands on F. For this growth pattern there are 4 notes and 48 beats, so each note will be 12 beats long. The long notes and fewer steps up communicate that pink did not grow much in the first 12 hours. Neon grows from D, D#, E, F – C. For this growth pattern there are eleven notes and 48 beats. Each note is 4.36 beats in length. So the first ten notes are four beats long, and the eleventh is eight beats. The longer note at the end places emphasis on the final growth number for that 12 hour period. Faster steps further up the scale sonify neon‘s more abundant 12 hour growth period.
Looking at the graph, it is easy to hear that the growth patterns of pink and neon invert at the 12-24 hour sampling unit. Pink leaps from .25 to .7, while neon short stretches from .6 to .75. Again, note duration and number of steps will sonify these contrasts in the data.
While a sense of growth is captured by the movement up the scale, there is not yet a sense of increasing density. To get at this, I decided to sustain the top note of each 12 hour sampling unit. As example, pink’s F and neon’s C would continue softly to the end of the 48 measures. This would follow for the last note of each 12 hour cycle and will create the sense of sonic density.
Enough talk, lets have a listen!
neon 48 hour growth pattern
pink 48 hour growth pattern
These are the 48 ms versions of the patterns. So 48 4/4 measures at 120 BPM really stretches out these relationships making it harder to hear the movement of the data. Ableton Live has a function that allows me to collapse the sequence from 48 measures to 24 measures and still maintain the rhythmic integrity of the phrase. WoW! Then the phrase can collapse to 12 measures. All of these phrases will likely be a part of the Sourdough Song, but I am still deciding which version (24ms or 12ms) conveys the data more clearly. One of the researchers on the project said the longer growth articulations conveyed the anticipation the bakers feel as they wait for their starters to grow.
Here is the 12 ms version of both strains together. See if you can hear the changes described above. Listen closely for each voice – you will hear pink holding longer tones, while neon changes tone more quickly. It helps to look at the graph while you listen.
This will likely be one a motif within The Song of Sourdough.
Data sonification is a burdgeoning area of sound design that is quite amazing in its depth and flexibility. I have a keen interest to sonify data in a way that furthers our understanding of the data. I would love to create a sonic pie chart for example. While a visual pie chart is a snapshot, a sonic pie chart would be more like an animation. A chemical reaction could be sonified by assigning particular voices to different parameters of the reaction: as the reaction proceeds, the voices would change from “reagent” voices to “product” voices. Consonance and dissonance couid illustrate the changing relationships amongst the components of the chemical reaction. One possible way to sonify, in my mind.
Then at Moogfest 2018, a workshop introduced me to the world of SuperCollider and MaxMSP as instruments for creating sonic pie charts. Mark Ballora of Penn State University (Please check out his work at http://www.markballora.com) has been working with sonifying data for decades. He was doing it when no one was paying attention. Mark uses SuperCollider to create sonifications of tidal changes and the movement of hurricanes. This type of sonic representation of data illustrates how a group of parameters changes over time, and when you listen, you hear all of the changes happening over time. Voila! A sonic pie chart! Attending Mark’s workshop, shifted my soundsense, as I realized I do not want to learn computer programming (at this time). This blog post by Mark Ballaro and George Smoot (https://www.huffingtonpost.com/mark-ballora/sound-the-music-universe_b_2745188.html) helped me understand that my interest is in exploring how modal/timbral shifts that are set in a familiar,equal-tempered scale spectrum might illustrate data-driven relationships. What I am interested in is more a sonic illustration, than a map or a pie chart.
Just before Moogfest, The Dance DL, a Durham dance listserve sent this announcement:
Rob Dunn’s lab at NC State University explores microbiomes of some of our most familiar places. The sourdough project studies sourdough starters from around the world, including some really ancient ones that have been passed down for generations. Seeking an artist working in any media with an interest in microbiology, bread baking, making the invisible visible, and/or communicating complex science through art. Help us bring the awe and wonder of science–and the microbial world– to the world.
As I read this notice, it felt like a dream! I have a two and half year old sourdough starter which is used to create 75% of the bread Trudie and I eat. I have recently studied cell biology, neurobiology and have a deep interest in molecular chemistry about which I am just learning. And I am looking for a data sonification project. I sent them an inquiry, they checked out my sound work, and I was invited to participate.
First step, meet with the Sourdough folks at Rob Dunn’s Lab. On Friday June 15th, Erin McKenney, post-Doctoral Fellow in Microbiome Research and Education and a research lead on the sourdough project, and Lauren Nichols, Dunn Lab Manager, met me in the lobby of the David Clark Labs (home of the Dunn Lab). I learned that the sourdough project is looking at the ecology of sourdough starter communities as relates to yeast and bacteria growth in flour when exposed to water and the local microbial environment. I attended a lab staff meeting and learned about the amazing research being done here. All the projects are basically looking at how the smallest phenomena impact much larger phenomena and vice versa, the micro to macro to micro feedback loop. And they keep finding that diversity is the key to sustainable growth and a healthy environment. I left the meeting excited and inspired! Next stop will be the As If Center in Penland, NC in October.
The only other preparation I would like to do is to try sonifying some data. I reached out to the Rob Dunn Lab folks, and Erin McKenney sent me a data set to try my hand at. The data is about nine lemur babies from three lemur species, and how the microbial makeup in each baby’s stomach evolves as changes are introduced to their diets. (This is Erin’s dissertation study!) We have identifiable parameters that can be orchestrated to show changes over time. Perfect!
The data is on a massive (to me) spreadsheet with lots of terminology I don’t know…yet. This will be an interesting process as we work out exactly what the sonic map will depict. I sense that certain data will lend itself to sonification and that is the part I do not yet know. After spending some time studying the spreadsheet, I asked Erin how we can cluster some of the microbial data together, and she sent me the class and phylum data sheets. Phylum became my focus as there were only 35 phylum as opposed to 95 classes and 255 strains of bacteria. One of the lemur mothers had triplets so I decided to put together phylum profiles on this small group. Culling through the data for these specific individuals narrowed the phyla divisions down to 24, then I made an arbitrary cutoff point of >.00 density for each phylum (Erin said this was fine and is actually a tool scientists use to declutter data). Now was down to 15 phylum – a manageable number for a timbral illustration.
The microbes were collected from the three babies six times from birth to nine months. The timeline for the samples was birth, nursing, introductory solid foods, regular solid foods, and two times as they were weaning. Microbes were collected from the mother when she gave birth. Erin had the brilliant idea to have the mother’s phylum profile (which does not change over time) be a drone under the babies’ phylum profiles in the sound map. This allows you to hear when the profiles diverge and when they converge.
The sonic substance for all this is a phyla megachord that stretches from G1 to G5. Each phylum is voiced by a single pitch, so, for example, Protobacteria is G1. Since there are only thirteen pitches in a chromatic scale, some of the phyla would land on the same pitch, different octaves. There were five phylum that tended to have the highest presence in each sample, so I made them the Gs, and all the rest had separate, distinct pitches. I used amplitude to render the amount each phylum was present in each sample.
Then there was how to voice the individual profiles in order to hear the data as clearly as possible. After much experimentation the mother’s voice is a woodwind with steady tone throughout. I chose bell-like voices for the three lemur baby profiles, letting each phase ring out four times over the mother’s profile. The idea is to listen and compare the mother’s profile with the babies’ profiles. Listen for the change (or lack of change) as the each stage rings in four times. You will probably need to listen closely several times. What you hear is a uniformity of tone at birth that becomes more dense and dissonant as the phyla diversify with the babies’ diversifying diet. Then the final wean profiles settle into more consonance with the mother’s profile. So very interesting!
When I sent this to Erin, she said, “The patterns you’ve detected and sonified are exactly what I published.” Yes! This is the sketch I will use to create a soundscape of the Lemur Data. From this exercise, some tentative questions have emerged that will help when we start working on the sourdough project:
How is the data organized/catagorized?
What is being measured?
What are the signifigant changes and time frames within the data collection process?
What are the researchers interested in hearing from the data?