Hiya bolsters AI Voice Detection with targeted treatment for background sounds

In an era where AI-generated voices can mimic anyone with uncanny accuracy, the stakes for mobile carriers have never been higher. Fraudsters are leveraging this technology to clone voices and deceive users, such as in this notorious Taylor Swift voice scam that targeted fans with a fake cookware giveaway.

These threats are made even more challenging by the messy, real-world conditions that exist in the exact places where these attacks take place — like background noise, music, and multiple conversations that often occur in online videos, social media, and phone calls. This is why Hiya has recently bolstered its industry-leading AI Voice Detection with targeted defenses aimed specifically at these challenges.

The real-world challenge: Background sounds

For most AI voice detection solutions, the leap from a controlled environment to the chaos of the real world is a chasm too wide. Real-world conditions disrupt accuracy, leaving many tools ineffective when it matters most.

Background sounds — such as music, conversations, and ambient noises — are a prime example of this. These are extremely common in online videos, phone calls, and video conferencing. And yet, it can be difficult for many solutions to perform well in these environments.

This is why Hiya recently added targeted defenses tailored to the challenges presented by background sounds. These targeted defenses help maintain Hiya’s industry-leading accuracy and stay one step ahead of the latest deepfake attacks.

How Hiya improved detection and treatment of background sounds

Adding to the high accuracy of our models, Hiya has introduced two targeted defenses aimed specifically at handling background sounds:

• Training data - The Hiya team has a unique understanding of the real-world conditions in calls, because voice is all we do. Using our specialized knowledge of the space — including knowledge gleaned from our global honeypots — we developed our own proprietary toolkits to generate training data with background sounds. This expands the training data that models are exposed to and ensures they are specifically trained against a wide range of background sounds. By generating our own toolkits for this, we can ensure models meet a large variety of real-world scenarios in training, and that happens at scale.
  
  

• Targeted modeling - We added new modeling aimed specifically at measuring and dynamically handling the impact of background sounds in a given situation. This means our models can assess whether audio contains background sounds, how disruptive those sounds may be, and which models need to be deployed (and how) in that specific situation.

These defenses join our existing ones to ensure the highest accuracy in situations with background sounds. For example, here are our results with the Taylor Swift scam mentioned above:

Hiya AI Voice Detection: A leading solution

Hiya’s AI Voice Detection solution uses cutting-edge deep learning models trained on thousands of hours of audio, analyzing and identifying the subtle artifacts that distinguish genuine voices from AI-generated ones. This innovation achieves:

• High accuracy: With over 99% accuracy vs. the most complex “in-the-wild” datasets, Hiya’s solution has strong performance in both controlled and real-world scenarios.
• Minimal audio requirements: Detection begins with just one second of speech, delivering results with near-instant speed.
• Language and channel independence: Hiya’s AI Voice Detection is built not as a point solution, but as a broad technology that provides protection no matter which synthesis tool or language is used.

Not only is the solution adaptive and highly accurate, but Hiya constantly monitors the AI voice generation space to find new areas of innovation and improvement. Our targeted treatment of background sounds is just one of many improvements we are constantly making to ensure our solution performs for you.

Why it matters for mobile carriers

As mobile carriers strive to enhance user trust and safeguard communications, Hiya’s AI Voice Detection provides unmatched accuracy and flexibility. Available via cloud-based APIs or on-premises deployment, this solution is scalable for any carrier’s needs. Additionally, its multi-channel compatibility ensures it performs seamlessly across telephone and digital audio formats.

While other solutions may excel in a controlled lab environment, Hiya’s tools are battle-tested in real-world conditions, making them the gold standard for mobile carriers looking to protect their networks from voice-based scams.

Join the fight against AI voice scams

Mobile carriers have a pivotal role in combating voice-based fraud. By partnering with Hiya, you gain access to industry-leading technology that not only detects threats but stays ahead of the latest AI advancements. Secure your network and protect your users with Hiya AI Voice Detection—because in the fight against fraud, staying one step ahead is the only option.

Want to experience Hiya’s solution firsthand? Download the Hiya Deepfake Voice Detector Chrome Extension or learn more about our product on our AI Voice Detection page.