Hybrid-Net is a real-time audio source separation tool that can generate lyrics, chords, and beats. The innovative transformer-based hybrid multimodal model combines various transformer models to tackle different challenges in music information retrieval. This AI-powered project focuses on generating chords, beats, lyrics, melody, and tabs for any song. The U-Net network model, along with Pitch-Net, Beat-Net, Chord-Net, and Segment-Net, based on the transformer model, work together to establish correlations between different elements in a song. The project also includes features like chord and beat detection, pitch tracking, music structure identification, and lyrics alignment. Supporting multiple languages, including English, Spanish, Portuguese, and more, AI Tabs can generate playable sheet music with editing capabilities. Find out more on the website.
https://github.com/DoMusic/Hybrid-Net