Toxic Detectron

• 

  Speech Classification Layout

Inspiration

With the development of Generative AI, we have gained some computation prowess. So, deploying this prowess for the betterment of the society is very fruitful. While playing games with friends or attending an event, unforeseen misconduct can happen anywhere. So with the help of Gen AI, we can detect this anomaly.

What it does

It actually detects the abuse or bad things in communication between two or more people. And Identifies the bad things in the conversation.

How we built it

We built it using Open AI Whisper for processing speech-to-text and a pre-trained BERT transformer (detoxify) to classify the type of abuse in the conversation.

Challenges we ran into

It was very hard to find the perfect model for this project, so we tried a few more models like HateBERT, but it didn't help. Also, It was challenging to communicate the data between the front end and back end of the application.

Accomplishments that we're proud of

At least, the model classifies what type of abuse is present in the text and speech.

What we learned

We learned to find the perfect model for this sort of task and also learned how to communicate speech between endpoints.

What's next for Toxic Detectron

Next, we can do fine-tuning of the model for the perfect prediction based on the purpose we wanted to do.

Built With

• css
• docker
• fastapi
• html
• javascript
• python
• transformers