Panel 1

Michael Nardone Remarks on ARLO
In these remarks, I hope to do a few things: to reflect on this past year of research activity and the confluence of this group of people who are all pursuing various aspects of recorded sound; to begin to frame the discussions that follow in terms of what kinds of research we might be able to produce through this confluence, individually and collaboratively; and, finally, to advocate for a kind of grey literature, a mode of technical reports that might facilitate a discussion of where we can go from here, what we can produce through our researches.

Eric Rettberg Looking for Laughter in PennSound
Given a tool that can search sound with other sounds, what kinds of sounds would we actually want to look for? While some have focused on the particularities of background noise or on the particularities of poets’ speaking voices, I’ll argue that the sound of laughter represents a compelling target. Laughter proves surprisingly central to the experience of reading or listening to poetry, and it’s a form of response that has gone largely unrecorded in the age of the book. By exploring laughter with ARLO, we can better understand authors’ irony and earnestness, find moments of distance between an author’s intentions and an audience’s response, and better understand the crucial role that laughter has played in twentieth- and twenty-first-century poetry.

Ken Sherwood Distanced Sounding: ARLO Visualization and Poetry Audio Versioning
One interest in spoken word poetry archives is the access to audio files documenting numerous performance instances of a given poem. How can audio visualization and tag discovery help to identify and represent degrees of variance between multiple instances? Which variant features (tempo, volume, pitch, rhythm …) are most salient for the human listener, and which can be identified through visualization and machine learning?

Panel 2

Juliana Nykolaiszyn In search of Oklahoma
Pulling from oral histories recorded with women who survived the Dust Bowl in the 1930s, this presentation will feature preliminary tagging analysis exploring “Oklahoma” and other key words in this oral history collection.

Elizabeth Roke ARLO for Archivists: The Potential of Machine Learning for Basic Metadata Generation
When collections containing unlabeled, unidentified audio recordings arrive in the archives, archivists generally must listen to each recording to provide descriptive metadata for researchers including date, content, and type of recording. As a result, many thousands of hours of audio in archival repositories remain unidentified due to the scarcity of staff resources and are hidden from researchers. What if we could automate part of this process of description? This presentation explores the potential for using machine learning to help archivists expose these hidden audio collections and make them more widely available.

Toneisha Taylor: Lomax, Cade and the Ethnography of Sound recordings: Investigation into the WPA Slave Narrative Collection
Since the beginning of the WPA Folklore project and the private funding of the collection of oral histories and personal narratives from formally enslaved people in the 1920’s and 1930’s there has been some controversy about the collection and preservation of sound recordings. This project recognizes the controversy around the original recordings and asks critical questions about preservation, access and possibility of use(s) for the recordings. John Lomax’s work within the FWP at the helm of the Folklore and Folkways collection projects has meant the preservation of significant recordings about Black life in the Deep South during and after slavery. John B Cade, a contemporary of Lomax, and African American historian collected over 400 interviews in 13 states while at Prairie View State College (now Prairie View A&M University). Where Lomax’s collection was largely publicly funded, many parts of Cade’s collection were privately funded even though he was employed by a public college. The proposed project is an critical ethnography of the archive and sound recordings. The questions asked reflect critically upon both the Lomax and Cade collections. Questions of race, class, and gender are asked in both the collection and preservation of the collections. While both deal extensively with the collection of narratives from former slaves the bodies that interviewed and collected these narratives, the intent and training of those engaged in collection, and the roles of those providing their narratives and personal histories matters in ways that impact current and past practices. The goal here is to use critical ethnography methods of collection and analysis to understand the ways we both silence and celebrate the voices within sound archives. ARLO could allow for the analysis of and comparisons between two very large corpora. It may be possible to identify locations of recordings, home locations of participants or duplications in narratives collected i.e. did the same person collect narratives for Cade and Lomax? Or, equally possible, did the same person give interviews to interviewers representing the projects directed by Lomax and Cade? What can we learn from a critical ethnography of sound when we compare these to large corpora.

Panel 3

Steve McLaughlin “Sound in Space: Crude Regression-Based Audio Classification”
What can we learn about a recording without listening to it? Can we automatically sort audio based on genre, gender, and speech style? Is it possible for an algorithm to judge the quality of a poetry reading? This paper will examine distant reading strategies for audio in general and speech in particular, with an emphasis on high-dimensional regression models.

Chris Mustazza: Forensic Audio Analysis to Determine the Provenance of Poetry Recordings
Is it possible to determine the provenance of poetry recordings based on artifacts from the materiality of their recording histories? For example, if we know that one set of poetry recordings was created on a specific recording device, can we use that recording to identify other recordings in an archive that were recorded on the same device, thus helping to determine their provenance?

Panel 4

Virginia Millington Accessing the StoryCorps Archive Using Digital Tools
With over 30,000 hours of audio in the StoryCorps Archive, the collection represents an ideal data set in which to test a diverse array of digital tools of discovery. From ARLO to OHMS, StoryCorps has pursued ways in which to search, access, analyze and share the contents of this remarkable resource, including investigating methods of discovery that not only highlight the words spoken in any given interview, but also its content, tone, affect, emotion, and more. This talk will include specific examples of this work, as well as a discussion of what the future holds for the StoryCorps Archive.

Tim Powell Tagging the Spirits: Using ARLO to Identify When the Spirit of the Drum Speaks to an Ojibwe Traditional Knowledge Keeper
Tanya Clement and David Tcheng used a video clip of Larry Aitken, an Ojibwe Traditional Knowledge Keeper, to demonstrate how ARLO can be used to identify different kinds of sound in the video– the drum beating, Larry speaking Ojibwe, and Larry speaking English. I have added another set of tags to try to locate when the spirit of the drum speaks and when Larry Aitken answers the drum back to try to reveal the spiritual dimensions of digital technology.

Kristen Saugee-Beaudry Ethical Considerations in Designing Digital Environments with Indigenous Communities
Traditional knowledge dissemination in Cherokee culture differs from practices that privilege artistic and intellectual freedom. Incorporating Cherokee consultants into all phases of design and development of online Cherokee knowledge repositories is crucial to successful and ethical scholarship.

Panel 5

Classification: feature discovery, the iterative approach, and looking at features in context (David Enstrom)
Clustering (Elizabeth Roke, Chris Mustazza, Michael Nardone)