Audio Data Process

Overview

There are two sources for audio on the DAILP site. This page describes audio recorded by DAILP during Zoom sessions. For audio recorded or uploaded by individual users, see User Contributed Audio

As of Spring 2023, DAILP has moved its long term storage and archiving of audio from from the Northeastern University Digital Scholarship Group's (DSG) Digital Repository Service (DRS) to DAILP-managed AWS S3 buckets in preparation for user submitted audio.

As of Spring 2024, DAILP has started moving all new audio recording to the DAILP website using the user contribution workflow.

Prior to Spring 2024, DAILP's audio data processes began data collection and recording over Zoom, followed by encoding and segmenting using Praat, and ending with archiving in the DSG Digital Repository Service (DRS). Archived audio segments were attached to entries in the DAILP database and displayed publicly on the DAILP website.

Before Recording

The document recording process can begin with any document DAILP has access to, provided the document has at least a syllabary transcription and a free translation. From here, there is a brief period where the translation team can highlight words or passages to present to our translation team during document recording sessions. Once these two steps are complete, a document can be queued for recording.

Data Collection and Recording

Currently, audio recording is performed using Zoom meeting recordings. Before beginning the recording call, Zoom is configured so that recordings are saved locally and, since we are not including video at this time, an audio only file is saved. Once saved locally from zoom, the files are converted to .WAV files to enter segmenting and encoding.

Audio files can be collected using other recording methods or collected from other sources so long as the files can be converted to the .WAV format for segmenting and encoding.

Segmenting and Encoding

Currently we are using Praat to segment audio by story and word. We use a TextGrid annotation with two layers, one for the story within a recording and one for glossing individual words. These timestamps will be used to associate segments of recordings with data in the DAILP database, and can be expanded to include morphemes, correlations to source syllabary text, and any other isolatable audio we may want in the future.

Archiving

All audio recordings of manuscripts and the corresponding annotations are saved in the DAILP Audio Manuscript Readings folder on the DRS. All files associated with a document should be in a collection named with this schema:

[Document ID]-[Document Title]

Additionally, metadata should be added describing the collection, including a document thumbnail and a brief description of the document, the recording process, and the annotation process.

Example:

Willie Jumper Manuscripts: Story of Sequoyah. 
4 Pages, page 1 and 4 read by Clara Proctor, page 2 read by Marlene Ballard, page 3 read by Oletta.
Pages 1 and 2 were recorded September 3, 2021; pages 2 and 4 were recorded September 17, 2021.
The first recording was annotated on September 9, 2021 and the second recording was annotated September 21, 2021.

Within each folder, there should be at least one audio file present. Each audio file should follow the naming scheme:

[Document ID]-[Document Title]-pp[Page number].wav

Metadata describing the recording medium and speaker, and the person responsible for recording should also be included. Additional information about the recording and annotation process is encouraged but not required.

Example Metadata:

Title:
  EFN02-Funeral Notice for a Child of Igakala
Creator:
  Proctor, Clara
  Trevino, Naomi
Date created:
  March 26, 2021
Type of resource:
  Sound recording
Format:
  Sound Recording
Abstract/Description:
  An audio recording of Echota Funeral Notice #2, Funeral Notice for a Child of Igaka'la (1925) in the DAILP collection. Recording was conducted over Zoom, recorded on a headset of unknown make and model in a room with 4 United Keetoowah Band community members. Recorded  by Naomi Trevino, Read by Clara Proctor
Subjects and keywords:
  Linguistics
  Cherokee

The annotation file should be named the same as the audio file entry:

[Document ID]-[Document Title]-pp[Page number]

Similarly, the metadata should reflect the annotator and contain relevant keywords.

Migration

Once a recording and its segmentation is loaded into DSG, it can enter our data migration process, where document-length audio will be associated with an AnnotatedDocument's meta field and audio segments will be associated with an AnnotatedForm (see the DAILP data rust documentation for more info on AnnotatedDocs and AnnotatedForms). From here, both sections of documents and whole documents can be played on the DAILP site, although this feature is still in development.