This week’s readings deal heavily with topics in oral history and the recording and collecting of digital audio. HiPSTAS, or High Performance Sound Technologies for Access and Scholarship, provide a toolkit useful to anyone compiling collections of audio files in creating sets of metadata and identifying speakers.
HiPSTAS has worked on a number of projects with the goal of developing different digital tool kits that can assist researchers in compiling, identifying, and defining segments of audio files. One such program is the Audio Labeler program, designed to label segments of an audio file usually in relation to a single speaker. The website goes into great detail about how you can train the program to identify the speaker you are focusing one by feeding it many small clips of that person speaking with that person, other speakers, other sounds and even silences already labeled. By doing this you can then train the program to apply that algorithm to unlabeled files so that it can label the files for you.
This tool would be very useful for compiling a collection of audio files on specific people, particularly for libraries and archives where there might be many different audio file collections. Identifying speakers and providing these labels is needed so that one who may be conducting research can search for these files. This program is meant to help expedite that process so that librarians or archivists or whoever is compiling these files doesn’t have to listen to every single audio file and individually label each segment.
It is not a system without some flaws though. The process involves feeding the program pre-labeled audio files, sometimes close a thousand of them. This means either someone still needs to individually label those small files or running a different program to do that, but those programs can at times be unreliable.
HiPSTAS’s most recent project has been to develop a formula to upload annotations to the segments of these audio files as well. The project, called AudiAnnotate, would allow researchers to translate their own analyses of the audio files into publishable media annotations. The hope for this project is that it would make audio file collections easier to search through and more accessible to researchers so that they are used more often and therefore preserved more often by libraries and archives.
Overall I found the projects that HiPSTA has and are continuing to develop very interesting. I hadn’t really thought about the problems of trying to search for a specific audio file before but these programs that make it easier to navigate audio files like their any other type of information seems really helpful for maintaining collections. I would be interested to see how AudiAnnotate works as its development progresses and how annotations of these files could be shared among researchers.