The problem DeepChain solves

With the evolution of Deep Learning, it has more becoming accessible and computationally inexpensive to make Deep Fakes, which could extremely dangerous going forward in the future. From sparking rage among communities to declaring wars, researchers are extremely worried about this technology. Enter DeepChain, a blockchain-based system that captures the source of truth and uses AI to speed up recognition. We first capture the original source into an immutable system, generating a unique token, which is embedded in the source. Then AI helps in parts like recognizing the token and also in speech transcription in the content and helps verify the validity of the information on various platforms( they can to live transcription and image recogntion ). It's can be trusted and efficient enough to be applied in the real world.

Challenges we ran into

There is two major challenge we bumped into :

The Google cloud's speech-to-text API requires a very specific format of audio encoding which was difficult to figure out as no clear documentation was given. After trial and error, we found that .wav format the mono channel sound works. As of now only .wav format are acceptable on our portal

We planned to implement a service where it takes a video, extracts out a frame, and run OCR to get the token. The team member was able to implement it successfully on his machine(running windows), But when we deployed it on a virtual machine (running Linux), getting the system to work with required libraries took too long, and yet the results were very inconsistent, sometimes returned complete blank response. In the end, we couldn't include in the project :(