---
license: mit
language:
- en
pretty_name: MelodyMaster V1 Training Dataset
task_categories:
- text-to-audio
- audio-classification
tags:
- audio
- music
- music-generation
- text-to-music
size_categories:
- n<1K
dataset_info:
  features:
  - name: ytid
    dtype: string
  - name: start_s
    dtype: int64
  - name: end_s
    dtype: int64
  - name: audioset_positive_labels
    dtype: string
  - name: aspect_list
    dtype: string
  - name: caption
    dtype: string
  - name: author_id
    dtype: int64
  - name: is_balanced_subset
    dtype: bool
  - name: is_audioset_eval
    dtype: bool
  splits:
  - name: train
    num_bytes: 53861
    num_examples: 100
  download_size: 31937
  dataset_size: 53861
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# MelodyMaster V1 Training Dataset

This dataset contains music-text pairs for fine-tuning MelodyMaster V1, an AI music generation model. Based on MusicCaps dataset.

## Dataset Description

### Data Format
- File Type: Parquet
- Contains music-text pairs for fine-tuning

### Data Structure
```csv
youtube_id,start_s,end_s,caption
```

- `youtube_id`: Identifier for the audio source
- `start_s`: Start time in seconds
- `end_s`: End time in seconds
- `caption`: Text description of the music

### Source
Derived from Google's MusicCaps dataset, curated specifically for MelodyMaster V1 training.