TL:DR
A platform for AI-generated boring podcasts to fall asleep to. Fully featured with multiple boring voice options and a customizable topic. We trained the LLM model ourselves.
How it works
- The user types a topic and chooses customizable voice options.
- This information is sent to our LLM model, which generates an extra boring script for the podcast based on the chosen topic
- A voice-over is created using the script and voice options selected.
- The audio is played back to the user until they fall asleep
How we built it
We used Next.js and React Native to build a dynamic frontend for our mobile app. Users can create accounts, log in/out, see the explore page, and generate their own sleep podcasts. We created the backend using MongoDB Atlas for the database and a Flask server, for user account authentication. We performed Parameter-Efficient Fine-Tuning (PEFT) on a Llama LLM using PyTorch and Hugging Face. We gathered the dataset of sleep podcasts by scraping RSS feeds and transcribing the .mp3 files using OpenAI's Whisper. We converted the resulting transcript into an .mp3 audio file and customized the voice using Deepgram.
What's next for NapCast
- Uploading generated podcasts to the platform to be shared
- Each podcast can be shared with a link
- Liking and disliking podcasts
- "trending" and "best of all time" categories
- video generation for the podcasts
- ads
- subscription features
Log in or sign up for Devpost to join the conversation.