{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/5a4f0bffb32d4d9b874194d81310b21c\" frameborder=\"0\" width=\"1664\" height=\"1248\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1248,"width":1664,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1248,"thumbnail_width":1664,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/5a4f0bffb32d4d9b874194d81310b21c-00001.gif","duration":561.7456120030001,"title":"Creating a Lo-Fi Bollywood Sample with AI","description":"In this video, I demonstrate how I used AI to generate a lo-fi Bollywood sample. I explore different music generation techniques using audio clips and text. I start with a recent paper called MusicGen by Facebook, which generates waveforms and allows conditioning on both text and melody. However, the initial model didn't capture the lo-fi element I was looking for, so I fine-tuned it using a dataset of 150 songs with lo-fi characteristics. I then add human intervention by editing the clip, adding reverb, slowing it down, and incorporating Eastern instrumentals. Finally, I experiment with extending the AI-generated track using text prompts. Although there are limitations, this approach shows promise for longer track generation and transitioning between themes. Code and notebook to follow."}