Software that works.
Audio worth exploring.

In a three-person collaboration on the UPF Sound and Music Computing master’s programme, I designed and implemented the acoustic augmentation pipeline: convolutional reverb, real ambient noise, random EQ and dynamic compression. I also contributed to training and evaluating a Conformer chord estimation model, using track-level splits to prevent data leakage.

The team evaluated robustness on MARL and Isophonics. The repository documents the method, results and limitations.

An independent research project at UPF. I recorded and engineered the dataset in Logic Pro, then trained two models over EnCodec tokens to compare continuous and categorical conditioning. I evaluated pitch tracking, interpolation and decoding strategies: lower validation loss does not guarantee more accurate musical control.

DAW Podcast Automation grew out of my own post-production work. It analyses tracks using RMS, LUFS and true peak, suggests markers and automation, and prepares preliminary mixes and masters against loudness targets. It works on copies of Logic Pro projects or audio folders. This personal project is in active development, not a finished product.