“Dear SEM”
Digital Aural Redlining
Dr. Nina Sun Eidsheim
University of California, Los Angeles
Photo Credit - William Matczynski
Introduction
How do listening biases spread to, manifest within, and proliferate throughout digital tools? Cross-feeding my own work on voice, race, and power with that of internet scholar Safiya Noble, I think about the digitized voice as an inflection of what she describes as “the power of algorithms in the age of neoliberalism and the ways those digital decisions reinforce oppressive social relationships and enact new modes of racial profiling” (2018, 1). Technological redlining is Noble’s term for this phenomenon, and I adapt it for my purposes as digital aural redlining. I use this term as a heuristic to better understand voice and listening culture(s) in a market and social economy that relies on digitized voices and listening.
In the world of COVID-19, many of us depend on digitized voices. How do we make voice technology more inclusive while recognizing the damage already done by digital aural redlining? To begin to answer this question, we could apply the methodology I call “listening to listening”; we might also note the ways in which both humans and machines perform power through listening practices (Eidsheim 2019, 27) or, in June Jordan’s unambiguous formulation, the ways “white power uses white English as a calculated, political display of power to control and eliminate the powerless” (1989, 29–40).
In listening to how voice technology listens, linguist John Baugh has noted an error rate differentiation of “45 percent of words from Black speakers and 23 percent from white speakers” for Apple products (cited in Link 2020). Allison Koenecke’s (forthcoming) team of mathematicians, engineers, computer scientists, and linguists suggests that this discrepancy is caused by training the software without sufficient audio data from non-white speakers. As I am a singer and humanities scholar, my research focuses on the deeper values which are performed through listening practices and which tacitly shape digital application development. Naming these elements can help to diagnose systemic issues in current voice-based technologies, and to counter the belief that digital environments may be more neutral than people.[1]
Digital Aural Redlining
Largely associated with the real estate and lending markets, redlining disproportionately saddles African American and Latinx customers with higher interest rates, fees, and banking premiums, putting them at an economic disadvantage (Biss 2015). In other words, the term describes practices that disadvantage individuals and communities based on race and class, regardless of individual character or credit score. “Aural redlining” describes a systematic listening practice that: first, others people based on their voices and on non-vocal contextual information; second, renders them either hyperaudible or inaudible; and third, due to the ubiquity of such digital voice and listening tools, disadvantages individuals economically.
Digital aural redlining produces a similar effect, in that voice technology discriminates against and segregates people through unequal access. “Digital” denotes listening practices that have been quantified in code and are used in technologies such as voice-to-text, virtual assistants, and automatic captioning.
The first type of digital aural redlining describes a situation in which listeners perceive a certain accent based on the ways they visually, verbally, or contextually understand the speaker’s race. Non-vocal cues affect how voices are perceived. For example, seventeen years ago singing voice synthesizer developer Vocaloid issued software which attempted to mimic soul singers, equating genre with race. To reinforce the connection, the company used racialized, minstrel-esque packaging featuring two figures with full pursed lips. Other contextual information about the voices, such as racialized descriptions, also framed the listening experience.
The second type of digital aural redlining describes the digital acoustic shadow[2] which, in effect, renders a person inaudible because their accent prevents or precludes them from effectively using many voice-based technologies. In court, transcription software is much more likely to lead to racial bias in automatic hate-speech detection models when applied to African American English as opposed to other English dialects (Marteen et al. 2018, 1668–78). Automated voice messaging systems, commonly used by banks, hospitals, and other vital services, feature a similarly discriminatory margin of error. The same technology is used in smartphones, making success in certain jobs more difficult for those whose accents are digitally discriminated against.
In the third type of digital aural redlining, voices are made hyperaudible through algorithmic identification of their alterity. We should be extremely concerned about and alert to unregulated surveillance and voice data collection, whether through personal devices such as phones, consumer technology such as Alexa, or systematic surveillance of minority populations such as China’s treatment of the Uyghurs.
Conclusion
As COVID-19 has accelerated reliance on digitized listening technology in a number of spheres from access to vital services to technology integration in the workplace, we must be vigilant in noting who gets lost in the acoustic shadow, and who is othered by hyperaudibility. Music scholars have a duty to lend our expertise in sound technology, voice, and listening to demand regulation in the digital realm in order to prevent social and economic discrimination. Part of the goal of my work, currently with the recently founded UCLA Practice-based Experimental Epistemology (PEER) Lab, is to illuminate these shadows and develop new ways of interacting with and creating solutions around voice.
Notes:
[1] For the power of naming, see Adler-Kassner and Wardle (2015).
[2] An acoustic shadow is a phenomenon in which—in the closest proximity to a sound source, or, if a solid obstacle obstructs sound waves—the sound is inaudible.
References
Adler-Kassner, Linda and Elizabeth Wardle, eds. 2015. Naming What We Know: Threshold Concepts of Writing Studies. Boulder: University Press of Colorado.
Biss, Eula. 2015. “White Debt: Reckoning with What Is Owed—and What Can Never Be Repaid—for Racial Privilege.” The New York Times, December 2, 2015.
Eidsheim, Nina Sun. 2019. The Race of Sound: Listening, Timbre, and Vocality in African American Music. Durham, NC: Duke University Press.
Jordan, June. 1989. “White English/Black English: The Politics of Translation.” In Moving Towards Home: Political Essays, 29–40. London: Virago.
Koenecke, Allison. Forthcoming. “Re-writing Algorithms for Just Recognition: from Digital Aural Redlining to Accent Activism.” In Thinking with an Accent, edited by Pooja
Rangan, Akshya Saxena, Ragini Tharoor Srinivasan, and Pavitra Sundar. Berkeley: University of California Press.
Link, Jeff. 2020. “Why Racial Bias Still Haunts Speech-Recognition AI.” Built In, July 26, 2020. Accessed January 13, 2021. https://builtin.com/artificial-intelligence/racial-bias-speech-recognition-systems.
Maarten, Sap, Dallas Card, Saadia Gabriel, Yejin Choi, Noah Smith. 2018. “The Risk of Racial Bias in Hate Speech Detection.” Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, 1668–1678.
Noble, Safiya Umoja. 2018. Algorithms of Oppression. New York: NYU Press.
Vocal Displacement, Adaptation, and Growth in the Context of Covid-19
Dr. Eve McPherson
Kent State University at Trumbull
Among my various roles, I teach voice at a community music school. One of my newest students recently immigrated from Cameroon. I have found the experience of teaching her very enjoyable, yet challenging, because our voices are rooted in different articulatory settings, what Beatrice Honikman identifies as the “gross oral posture and mechanics, both internal and external requisite as a framework…which constitutes the established pronunciation of a language” (Honikman 1964, 74). Articulatory settings are largely what create accents. Voice scholars also recognize that genre and style create unique singing “accents” (Miller 1997; Meizel 2020). As such, the forces of language, style, genre, and enculturation are all agents in creating the sound of the singing voice. My voice is rooted in bel canto technique and repertoire, so I feel vocally displaced when working with this student’s different vocal settings. She identifies her primary language as a form of pidgin English common to Cameroon, and when we work together, I lack the training in its phonation to have a complete understanding as to “where” she is coming from vocally. Since I do not understand her baseline articulatory settings, particularly the more difficult to observe internal muscle mechanics, I am unable to say exactly what position she should modify her voice “from.” I can only offer her a description of my suggested repositioning, which, with my training in bel canto techniques, gravitates towards dental consonants and Italianate vowels. Moreover, since we work entirely with facemasks on, I cannot see or demonstrate jaw, dental, and tongue positions as effectively as I might ordinarily. I am left trying to describe settings without the accompanying visual demonstrations. And, while these lessons have begun in a post-lockdown context, my sense of visual and linguistic displacement while teaching her has helped me to articulate my lockdown experience in terms of a variety of contextual vocal displacements.
A sense of displacement was very much present in remote lessons with students, their voices often eerily muted so that they could sing along synchronously with me. But a different type of displacement was also apparent in research I was conducting. Just before the lockdown, I had started a project on a sing-along event. Once each winter, the “Sing Along” brings together community members who select songs that are personally meaningful. They share their songs and stories and then the group communally sings the songs. The event, a meaningful ritual for its participants, is rooted in leftist ideology. Participants draw strength from the annual event to continue the exhausting work of community activism; they tell me they are specifically strengthened through entwined singing voices. The 2020 Sing Along was held in late January, but shortly thereafter, such gatherings were no longer possible. As I conducted my interviews throughout the lockdown, it was clear that participants keenly felt the group’s vocal displacement. Most understood that an in-person 2021 Sing Along would not happen and they felt a remote version would be unsatisfying. Still, they found a creative adaptation: a summer poetry reading conducted across backyard fences, voices working as recitation soloists. Participants vocally expressed communal values at a distance, a testament to the hearing of one another’s voices in the immediacy of the non-virtual world. (I find it interesting that they chose not to sing songs as soloists. From talking with participants, most feel too vulnerable when singing alone).
I personally experienced a different sort of vocal displacement. In the summer of 2020, my mother, who is an active amateur musician, suggested that we spend the July 4 holiday together making music. She sent along songs for me and my husband, an amateur guitarist, to learn. The songs were mostly American and Irish folk songs and mid-twentieth-century popular songs. With my pre-conditioned articulatory settings, the folk songs were already a bit of a stretch, but the popular songs really pushed genre boundaries for me. For years, I have felt uncomfortable singing genres that mix poorly with the bel canto voice – my default vocal setting. I learned this style from a very young age through participation in a respected Cleveland children’s church choir, which was led by conservatory-trained classical vocalists. Being in the choir was a form of vocal enculturation and, along with me, three other direct peers in the choir went on to formally train as professional opera singers, two of whom now teach in prestigious conservatories. So, for me the bel canto voice is very much a “natural” voice; I know this way of singing with little thought as to how I make it happen. In past literature, there has been an assumption that the bel canto voice is a contrived utterance; for example, in a preface to a collection of folk songs, Ruth Crawford Seeger once admonished, “The songs are better sung in a natural voice than in a bel canto voice” (Crawford Seeger 1941, xix). But I take issue with the notion of what a natural voice is from a personal perspective; contrived or not, bel canto is my natural sung voice. And I found that this vocal setting marked me as different in some respects.
An early memory I have of being vocally out of place was at the age of sixteen, driving with my friends to an amusement park with Blondie blasting on the radio. As I joyfully sang along with my friends to the songs, a friend began poking fun at my vocal sound and sang a Blondie song back to me with an exaggerated vibrato and a legato styling. I remember feeling deeply embarrassed. I hadn’t realized that I brought this “operatic” quality along with me wherever I went vocally. But there it was and, after that, with the typical insecurities of a teenager, I stopped participating in car sing-alongs. I also determined that I was not a good candidate for my high school’s show choir and never bothered to audition for it. Later, when I became interested in ethnomusicology, I again faced uncertainty as a singer studying the Turkish makam tradition. This experience of mismatched genre and voice type is something that I have explored as “timbral dissonance” in an earlier published essay (McPherson 2018). However, during the lockdown, in the privacy of my own home, as I worked through the songs my mother sent along for our pandemic-inspired family music ensemble, I let go of my inhibitions. I even began to sing rock and country tunes in the evenings with my husband, even though I brought along my “contrived” stylings as I sang. I overcame my sense of displacement and enjoyed singing without self-restriction or judgment.
Being vocally displaced has necessitated adaptations. I adapted the styles I sing so I could make music with others within the lockdown strictures of minimized social groups. I adapted my teaching techniques to the displacement of remote instruction. The activist singers adapted their means of vocal communion. These adaptations have inspired the re-evaluation of musical roles and the creation of new modes of connection. Up until this point, my research has focused on the acoustic and productive features of voice. I have also focused on individuals as soloists. Now, however, I am curious to understand more about the biology and psychology of group singing. Research has shown that choral groups begin to breathe and beat hearts almost as a unified organism (Vickhoff et al. 2013). I am now wondering how the structures of voices and unseen productive techniques begin to meld into one unit, even among casual singers, and how singing together shapes other community actions in various contexts. And, as to genre displacement, I will probably not study that, but I have added more joy to my music making, exuberantly, if privately, singing any song I want to sing.
References:
Crawford Seeger, Ruth. 1941. “Music Preface.” In Our Singing Country: A Second Volume of American Ballads and Folk Songs, xvii–xxiv. New York: MacMillan.
Honikman, Beatrice. 1964. “Articulatory Settings.” In In Honour of Daniel Jones, edited by David Abercombrie, Dennis Butler Fry, Peter MacCarthy, Norman Carson Scott, and J.L.M. Trim. London: Longman.
McPherson, Eve. 2018. “The Etic Voice: An Ethnomusicological Perspective on Voice Research in Turkish Secular and Sacred Practices.” In Singing: The Timeless Muse: Essays on the Human Voice, Singing, and Spirituality, edited by Darlene C. Wiley, 28–46. Gahanna, OH: Inside View Press.
Meizel, Katherine. 2020. Multivocality: Singing on the Borders of Identity. New York: Oxford University Press.
Miller, Richard. 1997. National Schools of Singing. Lanham, MD: Scarecrow Press.
Vickhoff, Björn, Helge Malmgren, Rickard Åström, Gunnar Nyberg, Seth-Reino Ekström, Mathias Engwall, Johan Snygg, Michael Nilsson, and Rebecka Jörnsten. 2013. “Music Structure Determines Heart Rate Variability of Singers.” Frontiers in Psychology 4 (1): 1–16.
Dangerous Voices
Dr. Katherine Meizel
Bowling Green State University
2020 was a year of dangerous voices. Voice, as a metaphor for agency, has long been inscribed with power that can threaten existing social structures. But beginning in March 2020, the material voice itself began to take on a shockingly literal association with danger. In one of the crueler effects of the pandemic, singing was discovered early on to be an act that spread aerosol droplets, and therefore viral particles, more quickly and across greater distances than speech. Gatherings of singers for rehearsals and performances became “super-spreader events.” By late March, U.S. news media were full of headlines about a Centers for Disease Control study regarding a choir rehearsal in Washington state that had led to a high rate of COVID-19 infection and two deaths (Hamner et al. 2020). In May the CDC first recommended that places of worship restrict singing—“Consider suspending or at least decreasing use of a choir/musical ensembles and congregant singing, chanting, or reciting during services or other programming”—though six days later this language was removed as the White House had not approved it (Chappell 2020). Certain sung languages with heavy plosive consonants, such as German, were identified through Japanese research as especially likely to pass along the virus (Craft 2021). And a May 2020 webinar organized by the National Association of Teachers of Singing, the American Choral Directors Association, Chorus America, Barbershop Harmony Society, and the Performing Arts Medical Association warned vocalists of the risks of in-person ensemble singing.
So, directors and vocal instructors scrambled to restructure lessons, rehearsals, and concerts in virtual formats—individually recorded voices edited together to create collective performances, in-car rehearsals in parking lots and driveways (Alone Together 2020; Merritt 2020). My choral director friends spent many extra hours carefully piecing students’ individually recorded voices together to try to erase time and distance, to establish an artificial sense of simultaneity and presence in the sound. Virtual choirs existed before the pandemic—besides Eric Whitacre’s famous project, now 10 years old, it is noteworthy that choirs of people with disabilities have also long utilized the same distance recording techniques that proliferated in lockdown (see Meizel 2020 for discussion of the Chronic Creatives Choir). The socializing benefits of group singing are well documented (e.g. Linneman et al. 2017), and research has long suggested health benefits as well (e.g. Więch et al. 2020). But in 2020, those positive effects were discursively and practically superseded by the idea of singers as disease vectors.
Still, it has been heartening to see how intensely and expansively creative singers have become in finding ways to make music together, at a distance. An explosion of growth in collaborative singing has spread across social media. TikTok, Instagram, and YouTube users have posted thousands of #singwithme challenges, in which an individual records one part of a song and invites followers to record and mix other parts. A cooperative musical based on Ratatouille, created through TikTok videos, garnered national attention and the eventual participation of Broadway stars in a new production; it was followed by another TikTok-based production built on the Netflix drama Bridgerton. Though many vocal groups have returned to in-person if masked rehearsals and performances, these projects remain part of social media culture as of the beginning of 2022. I can’t wait to see what the new year brings.
References:
New England Conference: The United Methodist Church. 2020. “Alone Together: Singing in the Parking Lot Choir.” Accessed January 1, 2022. https://www.neumc.org/newsdetail/parking-lot-choir-14899357.
Chappell, Bill. 2020. “CDC Quickly Changed Its Guidance on Limiting Choirs at Religious Services.” NPR, May 29, 2020. Accessed January 1, 2022. https://www.npr.org/sections/coronavirus-live-updates/2020/05/29/865324310/cdc-quickly-changed-its-guidance-on-limiting-choirs-at-religious-services.
Craft, Lucy. 2021. “Research Finds Singing in Some Languages Could Spread COVID-19 More Easily Than Others.” CBS News, January 28, 2021. Accessed January 1, 2022. https://www.cbsnews.com/news/japan-covid-19-singing-choir-research-languages-super-s
preader-coronavirus/.
Hamner, Lea, Polly Dubbel, Ian Capron, Andy Ross, Amber Jordan, Jaxon Lee, Joanne Lynn, Amelia Ball, Simranjit Narwal, Sam Russell, Dale Patric, and Howeard Leibrand. 2020. “High SARS-CoV-2 Attack Rate Following Exposure at a Choir Practice—Skagit County, Washington, March 2020.” CDC Morbidity and Mortality Weekly Report, May 12, 2020. Accessed January 1, 2022. https://www.cdc.gov/mmwr/volumes/69/wr/mm6919e6.htm.
Linnemann, Alexandra, Anna Schnersch, and Urs M. Nater. 2017. “Testing the Beneficial Effects of Singing in a Choir on Mood and Stress in a Longitudinal Study: The Role of Social Contacts.” Musicae Scientiae 2 (2): 195-212.
Meizel, Katherine. 2020. Multivocality: Singing on the Borders of Identity. New York: Oxford University Press.
Merritt, Rick. 2020. “Turn Your Radio On: NVIDIA Engineer Creates COVID-Safe Choirs in Cars,” NVIDIA, October 12, 2020. Accessed January 1, 2022. https://blogs.nvidia.com/blog/2020/10/12/covid-driveway-choir-denneys/.
Wiech, Pawel, Izabela Salacinska, Satarzyna Walat, Maria Kozka, Dariusz Bazalinski. 2020. “Can Singing in a Choir Be a Key Strategy for Lifelong Health? A Cross-Sectional Study.” Journal of Voice December 1, 2020. Accessed January 1, 2022. https://www.jvoice.org/article/S0892-1997(20)30423-9/fulltext#relatedArticles.