Yella offers a suite of products to secure phones against communication fraud. It includes tools like Sloka for cloned audio detection, Chitra for cloned pic detection, and TruSpeaker for on-device audio fraud detection. The product also includes a spam arrest filter for texts and Spam Caller ID to block spam calls using a frequently updated database.
Features
Sloka - Cloned Audio Detector
Flagship ZeroShot 100% accurate cloned audio detector that is server based, identifying and flagging artificially generated audio.
Chitra - Cloned Pic Detector
AI-powered server based tool designed to detect cloned pictures with 100% accuracy.
TruSpeaker
A 100% on-device iOS app for detecting cloned audio, ensuring real-time protection against audio fraud.
TruPhik
On-device iOS app for cloned picture detection ensuring privacy and security by processing data locally.
TruVideo
Real-time detection of cloned video and pictures on device, enhancing media authenticity.
Spam Arrest Filter
Utilizes LLM to filter spam text messages on-device, moving harmful messages directly into junk without notification.
Spam Caller ID
Uses a frequently updated FTC verified database to automatically identify and block spam calls in real-time.
TruSpeaker Compatibility
Supports various AI voice cloning tools.
Input Restrictions
Details any size or count restrictions for TruSpeaker.
AI Generator Detection
TruSpeaker flags certain AI generators.
Latency Information
Provides the latency details of TruSpeaker.
Privacy Protection
Discusses the privacy of audio recordings.
TruthPix Video Flagging
Can flag GenAI altered videos.
TruthPix Photo Flagging
Can flag GenAI altered pictures.
TruVideo Media Handling
Able to manage various types of media content.
TruVideo Restrictions
Restrictions on video length or size processed by TruVideo.
Deepfake Detection
This feature highlights the ability to flag deepfake audio that has bypassed YouTube's filters, showcasing the tool's accuracy and efficiency in detecting cloned audio.
Cloned Audio Identification
The tool can identify audio cloned by platforms like ElevenLabs and assess the authenticity of the speech, even in scenarios with artificial background noise.
High Accuracy Detection
The site claims a 100% accuracy rate in detecting deepfakes (0% EER), making it extremely reliable for users wanting to authenticate audio content.
Fake Image Detection
Spot fake digital images created or manipulated by AI technologies such as deepfakes.
Comparison Display
Display side-by-side comparisons of genuine and fake images for easy identification.
Cached Models
Loads ML model once on startup per thread, uses cached model on GPU, runs on A100 GPU in AWS. This improves scalability and performance by reducing model load times.
Real-time Processing
Processes audio and pictures in real-time; 15 ms for pictures and 81 ms for audio. This feature ensures quick processing with low latency.
Resource Efficiency
Uses less than 1GB RAM and scales linearly as the batch size increases. This optimizes resource usage, making it economical for larger processes.