Unlocking the Potential of AI
Trellis Data's AI Breakthroughs: Achieving 95% Accuracy Rate in Courtroom Transcriptions - Tim McLaren from Trellis Data Shares Insights on Ticker News
Hello and welcome back to Ticker News insight. I'm Benay Ozdemir. Artificial intelligence is dominating a change in the way businesses structure and model themselves. However, as data privacy becomes a hot topic for organizations, how can businesses overcome those challenges? Well, let's bring in Tim McLaren from Trellis Data to talk more about the delivery of a world leading result for the law enforcement industry through its cutting edge Trellis Intelligence platform. Thank you for your time.
Benay Ozdemir: AI is prolific at being used more and more by individuals and businesses. What are some of the ways Trellis data is seeing organizations take advantage of AI?
Tim McLaren: Thanks Benay, Look, AI's definitely moved from being a future tech to being a today tech, and our customers are realizing that they're going to have to deploy AI to make their processes more streamlined in order to stay ahead.
We've got offices in Australia and in the US and we work with some quite big organizations and government agencies and defense, and we're seeing A.I. being deployed in some really practical ways. In the logistics space we're seeing computer vision and scene text reading to read complicated and difficult to read identification on the side of containers and pallets and boxes to be able to to follow those goods through the system.
We're also helping the government to enhance their biosecurity screening at ports by looking for insects and pests on the sides of containers. In the defense space, we're helping to identify targets and listening for keywords and languages out in the field in the battlespace, and get that information back to the people who need it. And then obviously there's this speech to text. So this is where we're seeing quite a lot of growth, that idea of using natural language processing to hear a conversation, to transcribe it or translate it into to use that data as part of a business process.
Benay Ozdemir: Now, you've recently updated and you've announced these updates to platform designed to enhance the speech-to-text. What are some of the challenges in using AI for this purpose?
Tim McLaren: Yeah, that's a great question. So I mean, we're all quite familiar with speech-to-text. We tend to use it on our phone, but it comes with actually a lot of challenges. Within a single language, you've got different ways of speaking it. So formal or informal. Not all audio inputs are created equal. So the sound that's coming through on someone's mobile phone or their body cam or a walkie-talkie is going to be different.
Just how it's being spoken on the street and all those things can affect the model, the background noise. You've got grammar that you're trying to add over the top of everything.
So you know the context of how the conversation is playing out, words that are spoken in the courtroom might have a different meaning to those that are being spoken on the street. So the context is really important as well, because the model is always trying to predict what is going to be spoken next, identifying who's speaking and bringing that all together into a package that can be usable by the customer. And it's all challenges in there.
Benay Ozdemir: Now, how do you overcome those challenges and what are some of the ways that make the trellis data solution different?
Right. So we've got an in-house team of machine learning research experts. We call them ML Labs, and we really try and get to grips with the requirements of our customer and making sure that they that the model that we're building for them is going to produce the best results.
A model is only going to be as smart as the information that you feed it. So we work very closely with them to make sure that it's trained on language and data that's going to actually give you the best possible results. And we give them those tools to be able to enhance the model over time so that it can learn and get better as it is used and grows. We make sure that whatever the output that they use can actually integrate into the systems that they that they have.
In a recent tool that we've built for, for law, courts and courtroom settings, we achieved a 95% accuracy rate, which was above our competitors in that scenario. And you can imagine that's a huge advantage because our baseline platform can actually transcribe 600 hours of audio in a day. Normally would take a person 3 to 4 hours to transcribe just one hour of audio to 99% accuracy.
Benay Ozdemir: Now it seems that data and privacy big issues when it comes to your customers, having those concerns to.
Tim McLaren: Sure. So I mean, look, it's really important to understand how your data is being used and where
it's going when you're using AI tools. The adage is that if you're not paying for the service, you are the service. And there are some AI tools out there where you're not really ever sure where your data is going to show up.
In Trellis data, the platform that we've made is actually can actually be integrated into existing enterprise platforms. So it's a bit plug and play that way, which means it can live in a disconnected environment on premises. And the advantage there is that you understand that
where the data is coming from, from that, from the company's data source going into the model, being inference and the result coming back and then back to the user in a very secure way. And it means that they can meet all the security and privacy requirements of that organization because it's disconnected and it's on-premises.
Well, thank you so much for joining us, Tim.