The application is built on React Native + Python. 
It takes raw audio as an input and performs speaker diarization using pyannote.audio. Then by using Whisper it creates a transcription of the call. 
The transcribed text is summarized by GPT3 and analyzed by a blacklist algorithm that uses a list of words associated with popular scams. 
To improve algorithm performance we experimented with GPT, but only 3.5 version(chatGPT) was improving analysis quality. Since only GPT3 is available through API, we decided to wait with adding GPT to the algorithm.
Summarized text with a calculated probability of scams is being sent to user's relative.


🗓️ This will be a 7-days of hacking and fun from 9-16 December
💻 Build with the latest AI tools from OpenAI to create innovative new apps
💡 Work with top AI professionals and learn from them
⚒️ Create your AI app by combining GPT-3, Codex, Dalle-2, and Whisper
🐱‍💻 Register now and let's get started!

OpenAI Whisper, GPT3, Codex &  DALL-E 2 Hackathon

Fulstack dev + ML enthusiast. 
Working on an app that protects users from phone scams.

SafeCall mobile app