Advanced Seq2Seq Music Pipeline Using U-Net and TransformersAdvanced Seq2Seq Music Pipeline Using U-Net and Transformers
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Engineered a Sequence-to-Sequence (Seq2Seq) symbolic music generation pipeline utilizing U-Net, Bi-LSTM, and Encoder Transformers to synthesize simplified MIDI and sheet music, achieving an 89% F1-score in difficulty-level prediction.
Architected a scalable ETL pipeline to parse and process 440 complex temporal files from AWS S3, engineering a novel, time-aligned dataset specifically designed for difficulty-controlled Music Information Retrieval (MIR).
Developed temporal synchronization algorithms to align machine-transcribed audio stems with simplified symbolic sequences, establishing complex multi-modal mapping to train robust token-to-token translation models.
Post image
Hitesh's avatar
metrics and loss
Post image
Post image
Post image
Back to feed
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started