Serene Stream

• 

  In the other tab of the app are the tracks you've created! Connect with your community and find other similar tunes!

• 

  Generating a Song! This takes in an inspo prompt and a title.

• 

  Selection of clips for generating your own lofi or for scrapping.

• 

  Clips Page when you first enter. You can record sound clips for remixing

• 

  The statistics; just goes to show how much impact Serene Stream could have in the LGBTQ Community.

Inspiration

The LGBTQ community experiences disproportionately higher rates of depression, anxiety, and suicide. While the cause of this is systemic and something we're still working through as a society, the age-old balm to mental health has always been nature. We were also inspired by National Parks which allow people to support the preservation and ownership of natural treasures simply by being in the parks. We wanted to combine attachment to our local greenspaces with investment in one's own mental health through nature. Our app hopes to foster creativity in the outdoors without impeding on being present when there.

What it does

Serene Stream encourages you to engage with their surroundings by recording sound bytes: bird calls, water features, leaf rustling, etc. The app is minimalistic and audio-based to allow you to focus on their surroundings. After saving a few clips, these clips can be remixed with a chosen prompt into a personalized AI-generated lofi song. We hope the community can always bring a piece of nature home with them to build that connection and invest in their mental health. Furthermore, after creating tracks, we utilize an audio-to-vector comparison to fetch anonymous tracks from all over the world that are most similar to yours. In this way, Serene Stream fosters a sense of connection by nature lovers everywhere.

How we built it

We built the front-end on Swift, as it was important that our app was a mobile app. The backend API was built with Flask and the backend was on Vercel and MongoDB Atlas. Our GenAI lofi base used Suno.AI, which had music generation models. The sounds were remixed and overlayed using dark magic, and the sound-to-sound similarity comparison was done using a vectorization and audio2vec algorithm.

Challenges we ran into

The biggest challenge that we faced was with the Audio AI. We first needed a generated lofi base, then a method for remixing the collected sounds with that base. This proved to be pretty challenging since music generation software was mostly either proprietary or in a "train-it-yoself" academic paper. Furthermore, audio-to-audio models were also fairly uncharted territory.

Accomplishments that we're proud of

We are proud of venturing outside of our comfort zones. Nearly every aspect of stack was new to our team and we are proud of our adaptability!

What we learned

Working with audio and music generation in the context of AI showed us just how the field is still in its infancy as compared to visual and language generation. Many factors that make music so very human, make it incredibly difficult to capture and manipulate.

What's next for Serene Stream

Better incorporation of remixing and generation AI; we hope that creating music out in nature is something users can enjoy and improve through their careful selection.

Built With

• flask
• python
• suno
• swift
• vercel
• xcode