FL YuE2 · Prepare Dataset:
The FL_YuE2_PrepareDataset node is designed to facilitate the preparation of datasets for training purposes within the FL YuE2 framework. This node is particularly beneficial for users looking to align and prepare their audio datasets with corresponding lyrics and musical scores. It offers a streamlined process to ensure that all necessary components, such as vocals, lyrics, and musical scores, are properly aligned and ready for training. The node supports various configurations, allowing you to customize the preparation process according to your specific needs, such as aligning lyrics with recordings or planning scores for training. By automating these tasks, the node significantly reduces the manual effort required in dataset preparation, making it an essential tool for AI artists and developers working with audio datasets.
FL YuE2 · Prepare Dataset Input Parameters:
dataset
This parameter represents the dataset you wish to prepare. It is crucial as it forms the basis of the preparation process. The dataset should be in the YUE2_DATASET format, ensuring compatibility with the node's operations.
assets
Assets are additional resources required for the preparation process, such as training models and alignment tools. They are specified in the YUE2_TRAINING_ASSETS format and are essential for tasks like lyric alignment and score transcription.
align_lyrics
This boolean parameter determines whether the node should separate vocals and align reviewed lyrics to the recording. When set to True, it requires additional alignment models and increases preparation time. The default value is False.
cache_directory
This string parameter specifies the directory where prepared datasets will be stored. The default value is "prepared," and it is located inside the output/yue2_training directory. This parameter helps organize and manage the output files efficiently.
score_planning
This parameter allows you to choose the level of score planning for training. Options include "off," "melody," and "full." The default is "off," which means no score planning is applied. "Melody" expects a melody-only score, while "full" retains both melody and chords. This setting influences how the node processes musical scores during preparation.
transcribe_missing_scores
A boolean parameter that, when set to True, enables the use of the SheetSage2 tool to create missing ABC scores during preparation. This tool is licensed under CC BY-NC 4.0. The default value is False.
FL YuE2 · Prepare Dataset Output Parameters:
YUE2_PREPARED_DATASET
The output of this node is a prepared dataset in the YUE2_PREPARED_DATASET format. This dataset is ready for training and includes all necessary alignments and preparations as specified by the input parameters. It is a comprehensive package that ensures your data is optimally configured for subsequent training processes.
FL YuE2 · Prepare Dataset Usage Tips:
- Ensure that your dataset and assets are in the correct format (
YUE2_DATASETandYUE2_TRAINING_ASSETS) to avoid compatibility issues during preparation. - Utilize the
align_lyricsoption if your project requires precise synchronization between vocals and lyrics, but be prepared for longer processing times. - Consider using the
score_planningfeature to enhance your training data with musical scores, especially if your project involves melody or full musical compositions.
FL YuE2 · Prepare Dataset Common Errors and Solutions:
ValueError: Score planning must be off, melody or full
- Explanation: This error occurs when an invalid option is provided for the
score_planningparameter. - Solution: Ensure that the
score_planningparameter is set to one of the valid options: "off," "melody," or "full."
Missing alignment models
- Explanation: This error may occur if the
align_lyricsoption is enabled but the necessary alignment models are not available. - Solution: Verify that the required alignment models are included in the
assetsparameter and are correctly configured.
Directory creation failure
- Explanation: This error can happen if the specified
cache_directorycannot be created due to permission issues or invalid paths. - Solution: Check the path specified in the
cache_directoryparameter and ensure that the application has the necessary permissions to create directories in the specified location.
