Nick Rossow is the acting director for e-research services at Griffith University, and has helmed a project using Microsoft Azure cognitive services to automate transcription. The University had trialled several automated transcription technologies in the past – but none met the standards it needed either in terms of accuracy or proper information governance.
Griffith has built its system using Microsoft Azure Cognitive Services with additional support from Microsoft engineers. A transcription portal built using APIs to Griffith’s information systems is accessed by researchers. This essentially hides the complexities of the system from the researcher who accesses the system via Griffith’s identity management system to ensure proper information governance, accountability and security, then uploads the audio file, completes a couple of check boxes, and then downloads a copy of the transcription.
Rossow says the vision is for each research group to get its own deployment of that interface to the portal to help with expenditure traceability.
“Each project team has to pay for their own transcriptions. And the easiest way for us to manage that currently is to deploy a single instance of the service per project team. So each project team can log into their own service, and any transcriptions will be charged back to that project.”
The quality of transcriptions achieved to date has been reasonable and comparable to what can be achieved using other automated transcription services, according to both Rossow and Kendall. Griffith is now exploring ways of training the system to enhance accuracy, and personalising it to specific areas so it can handle complex jargon or expressions that are common in the rehabilitation or disability sector. They are also exploring ways of enhancing the quality of audio recordings and the protocols for researchers to use in managing recordings.
So far, the researchers like it. Rossow notes; “It’s in a pilot phase because we haven’t really started promoting it. While it’s not perfect YET, it already has over 20 users, which is great for word of mouth advertising.
“Right now, we’ve got some really good user feedback and need to do some more development work on the model to enhance accuracy. Training the model is a big focus for us at the moment, but also taking on some of the feedback about how we make the user interface more user-friendly”
He says that the increased focus on security and data governance resonates strongly with the researchers who feel more confident sending confidential voice files to the University’s Microsoft Transcription Service.
Rossow knows that the automated transcription service will be able to scale to meet researchers’ needs.
He says that; “No research group will impact another research group. If they decide to log in and throw thousands and thousands of hours of recordings at their transcription service, there’ll be no impact on the researcher sitting next to them using a completely different instance of the service. That’s a key benefit for us.”
And importantly, says Prof Kendall, the Microsoft technology promises to decimate the cost of transcription, allowing scarce research funds to be used in more efficient ways.
The ability to access transcripts much faster is also valuable she says, as it promotes continuity of the research rather than delays while waiting for access to the data for analysis.
“It definitely makes us more efficient, which is always worthwhile; but it also puts a lot more control back in the hands of the researchers, allowing them to manage their own workload and to work better in teams,” she says.
The service has been particularly valuable during the 2020 pandemic when researchers have been able to continue to work remotely, interview clients, arrange automated transcripts, and then have those in a central secure Azure based repository accessible only to authorised users.
According to Prof Kendall; “It just adds to the quality of the working environment – we’re all working online but we’re still working together and we can see how projects are progressing. We’re always looking for a better solution; to be able to do things quicker and more efficiently on public money.”