juancopi81's picture
Update README
37a7972 verified
metadata
dataset_info:
  features:
    - name: text
      dtype: string
  splits:
    - name: train
      num_bytes: 6417909784
      num_examples: 244436
    - name: test
      num_bytes: 1221971111
      num_examples: 46005
    - name: validation
      num_bytes: 1465985310
      num_examples: 54947
  download_size: 974110589
  dataset_size: 9105866205

Dataset Card for "lmd_clean_8bars_32th_resolution"

More Information needed

Available at Portex

🎵 Lakh MIDI to MMM-Style Text Dataset

This dataset converts the Lakh MIDI Dataset into a structured text format inspired by the Multitrack Music Machine (MMM) paper. It includes 344,900 samples, each representing an 8-bar symbolic music fragment, tokenized into a language-model-friendly format.

Each line in the dataset is a music fragment composed of tokens like:

PIECE_START COMPOSER=JOHN_FARNHAM PERIOD= GENRE=TIME_SIG=4/4 TRACK_START INST=122 DENSITY=0 BAR_START TIME_DELTA=48 BAR_END ...

🔍 Metadata

  • Modality: Text (converted from MIDI)
  • Format: One tokenized sequence per line (plain text)
  • Size: 344,900 rows
  • Source: Derived from the Lakh MIDI Dataset
  • Structure: Each row represents an 8-bar segment tokenized to match MMM syntax

🤖 Use Cases

  • Pretraining or finetuning symbolic music models
  • Sequence modeling research for music
  • Input for generative transformer models
  • Creative AI applications in music composition

🧠 Why this dataset?

Symbolic music datasets in tokenized, language-model-ready formats are rare. This dataset bridges audio-derived symbolic data and the world of NLP modeling, saving hours of preprocessing and formatting work for researchers and ML developers.