VoiceGuard

Inspiration

We learned about the scams going around the internet where someone mimics another person's voice and conceals their voice to manipulate others into giving them money. This includes masking oneself as a certain celebrity to market a false product. We have an ambition to protect innocent people unaware of this, especially the elderly and young. We hope to save large quantities of money from being stolen from people unwillingly with such tactics.

What it does

VoiceGuard can detect AI voices from a real human voice. This can protect you from scams that try to imitate another person's voice to deceive you into giving them money. It works quickly and only needs a quick file upload onto the app via audio file. We prioritized accuracy and ease of use when developing this project.

How we built it

We used certain speech patterns and tones exclusive to AI audio that can be filtered to detect false voices as to real ones. We displayed the AI’s resullt/response by displaying it on the UI. The AI is capable of developing a knowledge of these patterns over time to improve itself as well for future version tests.

Challenges we ran into

At first, we had difficulty implementing the AI to work with audio while allowing it to process it through its filtering system. Bug testing, implementing API’s, and troubleshooting fixed this. Later challenges were properly getting the AI to do its job with accuracy. There were often false positives or false negatives, and it was difficult to improve it over time. However, using testing vs. predicting variables, we were eventually able to lower the error margin substantially.

Accomplishments that we're proud of

We are proud of being able to train an AI to accurately detect false voices. We are happy to make a program with the potential to save large amounts of money being stolen. In more specific accomplishments, we were happy to overcome our programming/design challenges and successfully implement an AI in a topic we less commonly do in programming: to detect other AI in audio.

What we learned

We learned the difficulty of training an AI intensively under (image/visual) related conditions, as well as the difficulty of implementing an AI in that category of data. Exploring API’s to get over programming obstacles were also a skill we developed, as we were often stuck without sufficient functions to complete our tasks.

What's next for VoiceGuard

We hope to improve the AI's capabilities and improve the user interface. We also hope to incorporate user feedback to build a stronger user-developer connection. With its spread, it could possibly save incredible amounts of money.

Built With

• ai
• python