FraudCallDetection application

FraudCallDetection application

FraudCallDetection is an application designed to identify and flag fraudulent phone calls, with a strong focus on detecting deepfake audio using advanced data analysis and machine learning techniques

Created on 9th March 2025

FraudCallDetection application

FraudCallDetection application

FraudCallDetection is an application designed to identify and flag fraudulent phone calls, with a strong focus on detecting deepfake audio using advanced data analysis and machine learning techniques

The problem FraudCallDetection application solves

FraudCallDetection addresses the escalating issue of phone-based scams, which have evolved from simple spoofed numbers to sophisticated AI-driven frauds like deepfake audio impersonation. In an era where trust in phone communication is eroding due to relentless scam attempts, this app provides a robust shield for users across various scenarios, making call-related tasks significantly safer, easier, and more reliable. Here’s how it delivers value:
Thwarts Financial Scams: By identifying fraudulent calls—such as those posing as banks, tax authorities, or tech support—it prevents users from falling victim to phishing schemes that lead to monetary loss or stolen credentials. For instance, a scammer using a deepfake voice to mimic a bank representative can be flagged before sensitive information is shared.
Combats Deepfake Threats: With AI-generated audio becoming a tool for scammers to impersonate loved ones or authority figures (e.g., a fake "grandchild" in distress), FraudCallDetection’s deepfake detection fills a critical gap. It analyzes audio patterns to catch manipulations that human ears might miss, protecting users from emotional manipulation or coercion.
Streamlines Call Management: Manually screening calls is time-consuming and error-prone, especially with the volume of robocalls and scams today. This app automates the process, instantly analyzing metadata (like call frequency or spoofed IDs) and audio in real-time, so users can focus on legitimate conversations without constant suspicion.
Safeguards Vulnerable Populations: The elderly, non-tech-savvy individuals, or those unfamiliar with scam tactics are frequent targets. FraudCallDetection offers an intuitive safety net with instant alerts and customizable rules (e.g., blocking calls from certain patterns), reducing their risk without requiring technical expertise.
Boosts Confidence in Communication: For professionals who rely on phone calls—think freelancers, small business owners, or remote workers.

Challenges we ran into

Building FraudCallDetection came with challenges, particularly integrating deepfake detection with real-time call analysis. Here’s how we overcame them:

Sparse Deepfake Training Data: Early detection struggled due to limited AI-generated call samples. We solved this by generating synthetic deepfake audio with WaveNet and Lyrebird, improving training data and detection accuracy.

Real-Time Processing Latency: Initial processing delays of up to 3 seconds made the app sluggish. We optimized audio feature extraction using a lighter MFCC pipeline and multi-threading, reducing latency to under 500 milliseconds.

False Positives: Overly sensitive models flagged legitimate calls, frustrating users. We refined classification with a confidence threshold and user feedback, improving accuracy.

Platform Compatibility: Crashes on some Android devices stemmed from inconsistent audio APIs. Standardizing audio capture with a cross-platform library and adding fallback mechanisms ensured reliability.

Scalability Under High Call Volume: Heavy call loads strained memory management. During stress tests, the app buckled when processing multiple simultaneous calls, a scenario common for power users like call centers. The bottleneck was in memory management for the ML model. We optimized it by implementing batch processing for metadata analysis and offloading heavy deepfake detection to a cloud API when local resources were maxed out, maintaining performance under load.

Each challenge refined our approach, making FraudCallDetection more robust in tackling real-world phone fraud threats.

Tracks Applied (1)

Ethereum Track

FraudCallDetection leverages advanced machine learning to detect fraudulent phone calls, including deepfake audio, align...Read More
ETHIndia

ETHIndia

Discussion

Builders also viewed

See more projects on Devfolio