This post has been republished via RSS; it originally appeared at: New blog articles in Microsoft Community Hub.
Having worked in voice technology for most of my career, I can't help finding it ironic that on my smartphone the only app that doesn't ever appear to change is...well...the phone. But just when I thought the phone call was going to stay the same forever, the game changes entirely thanks to Artificial Intelligence (AI). The increased adoption of cloud by telcos coupled with the incredible advances of generative AI in the past few years provides an exciting opportunity for the telco industry to innovate on voice like never before, and have a meaningful impact on the life of their customers and a positive impact for society.
One area that has been top of mind for me and many of those I talk to in the industry is phone scams and fraud. In spite of the tremendous advances in technology in this area over recent years, telephone scams continue to be an enormous and, sadly, growing issue. The FTC reported $850 million lost to telephone fraud in the United States in 2023[1], a rise of 7% from 2022, with an average of $1500 lost by each person. Behind those numbers are many human stories of lost savings and impacted lives, and I am not alone in saying people in my life have been directly affected. I am therefore proud of the public preview of Azure Operator Call Protection. Call Protection is another powerful tool in the toolbox for fighting consumer fraud. It is a new service that applies AI analysis to phone calls in real time, alerts the subscriber if the call is suspicious and educates them as to why, for example a call supposedly from a bank that requests a PIN, something a genuine bank would never do.
So how does it work? What I am about to describe applies not just to Azure Operator Call Protection, but to applying AI to telephone calls in general.
The approach we have taken is to deliver the service from the network itself, as opposed to the handset, playing to the strengths of our operator partners in a number of ways.
- It allows operators to have control over how the service is packaged and delivered to their customers. Operators own and highly value the relationship with their subscribers and are best placed to enable the service for their customers.
- The service is available between any devices, whether cell phone or landline. This extends access to the widest range of phone users, including older subscribers who may be more vulnerable to telephone fraud, while also wedded to their landline.
- No handset app download is required. While providing an app can help enable an enhanced customer experience, it also requires the subscriber to discover and download the app, which can severely limit subscriber uptake. By providing an entirely network-derived experience, with no app needed, the service is available to as many subscribers as possible.
Two technical components are needed to apply AI to telephone calls:
- a gateway to get the call into the cloud,
- the AI services themselves.
Azure Communications Gateway is such a gateway and is an essential piece of technology for any modern operator looking to innovate in their voice network. It securely integrates operator voice networks with Azure and cloud communications services including Microsoft Teams, Azure Communication Services, Dynamics 365, and Zoom Phone. And now it integrates with AI services as well. Calls are hooked into Azure Communications Gateway using industry standard mechanisms such as IFCs (Initial Filter Criteria) and then sent over standards-based interfaces into the cloud. Once integrated into the network, it is easy to turn up additional cloud-based services. In other words, once the first AI-enabled voice service is up and running, it is a much smaller step to turn up service number two.
So now the phone call is anchored into the cloud, how are AI services applied in real time?
The first step is to use Azure services to transcribe the call to text. Speech-to-Text technology has been available for years, and I use it to dictate text messages on my phone (to the horror of my teenage children!). The quality of the transcription is high and continues to improve.
Secondly, take the call content now in hand and apply the desired AI techniques appropriate for the application. This may be to detect a potentially fraudulent call, or to translate into a different language or to answer a query. It takes careful engineering both to make the most cost-effective use of the AI, but also to achieve the most accurate results. We take privacy of customer and end-user data extremely seriously and have engineered Azure Operator Call Protection from the ground-up to ensure that it has all of the privacy safeguards that you’d expect from a telco-grade Microsoft product. The use of the AI is guided by a core set of responsible AI principles: fairness, reliability and safety, privacy and security, inclusiveness, transparency and accountability.
And, finally, take the output from the AI engine, and then use whatever techniques are appropriate to deliver the value back to the subscriber. For example, use Text-to-Speech technology to render the text back into audio, in order (for example) to warn the subscriber that the call is potentially fraudulent.
That’s all very well, but is it economical? Costs of AI technology have gone down enough now that it is possible to build cost-effective AI-based voice services, but as mentioned earlier, specialist engineering is required to use AI efficiently.
When talking to operator partners, one of the key concerns I hear is the desire to differentiate from other competing operators in region. How can an operator take a general purpose service like Azure Operator Call Protection and provide something different to their customers? Great question! The answer is in the unique data that is held within each individual network and figuring out how to combine that data with the Azure for Operators products to create enhanced value for customers. For example, in the phone scam scenario, many operators already have scam analytics or caller authentication (STIR / SHAKEN) information based upon call data from the network, which the operator could choose to combine with Call Protection in order to allow operational staff to better protect their customers by identifying patterns of scam activity or pinpointing sources of scams more accurately.
Azure Operator Call Protection is a great service that focuses on the safety of operator consumer customers. Where could these AI-powered services go next? Imagine the following, for example.
- A new arrival in a country has access to live language translation on the phone, allowing them to more easily talk to local organizations such as government agencies, or doctor’s offices, enabling a smoother transition to their new home country.
- An electrician who is frequently on the road has an AI assistant on the phone that automatically takes notes, schedules appointments, provides reminders and answers questions.
- Having the call content in text form in hand is a stepping stone towards being able to provide an amazing voice call experience for subscribers with accessibility needs – for example assisting those who are hard-of-hearing.
This is one situation where I am glad to have been wrong. Once again this is an exciting time for the telephone call. The possibilities for innovating are endless.
[1] As Nationwide Fraud Losses Top $10 Billion in 2023, FTC Steps Up Efforts to Protect the Public | Federal Trade Commission