Senior Music Ml Engineer

We are looking for a Senior Music ML Engineer for an R&D project.You will work with open data as well as the company's own corpus.
You will lead the research and development of the project together with a multidisciplinary team.We are not looking for an AI Engineer who deploys, integrates or prompts LLMs.
We are looking for someone who knows models from the inside and can modify and train them on a variety of music encodings.What you'll do Design the representation and tokenization of the notation.Understand and model the dimensions we want to be able to navigate in latent space.Train and fine-tune autoregressive models on symbolic notation, conditioned on metadata and several training signals.Define the evaluation: what are we modelling, and how do we evaluate it?
Prepare the model for inference together with the development team.Document experiments and results reproducibly.Requirements More than 5 years in machine learning, at least 2–3 of them in MIR and symbolic music representations.Hands-on experience training sequence models on music data, not only on text or audio.Command of notation formats (MusicXML, ABC, MIDI, Humdrum **kern, MEI) and of the tooling ecosystem: music21, partitura, MidiTok or Verovio.Advanced Python and PyTorch.
You have implemented or modified transformer architectures (attention, positional encodings, embeddings, etc.).
Familiarity with deep learning environemnts such as Hugging Face (Transformers, TRL, PEFT).
Hands-on Multi-GPU training (DDP, FSDP or DeepSpeed) and experiment tracking.Solid musical training: fluent score reading and a strong grasp of harmony, theory, and sheet music syntax, grammar, and engraving.
You need to be able to see why a model output is musically right or wrong in the modelled dimensions, perceptually.Knowledge of the state of the art in symbolic generation.Bachelor's or master's degree in Engineering, Computer Science, Sound and Music Computing, or equivalent experience.Language: fluent technical English.
Catalan or Spanish.Nice to have Publications at ISMIR, TISMIR, ICASSP, or the music workshops at NeurIPS/ICML.Open-source contributions to the MIR ecosystem.Experience in OMR.Deterministic proposals or ideas for solving the problem.
Not everything has to be solved with autoregressive models.What we offer 12-month full-time contract.A music notation corpus that is unique in the world.Cloud GPU compute for training and inference.Direct collaboration with a team of musicologists and music editors.
#J-*****-Ljbffr