meetween/mumospee_libritts

---
license: cc-by-4.0
language:
- en
---
### Dataset Summary
This dataset is a derived version of the [LibriTTS](https://openslr.org/60) corpus, converted into larger parquet files for optimized I/O performance on high-performance computing clusters. The dataset maintains the high-quality, multi-speaker, text-to-speech (TTS) alignment of LibriTTS, with over 585 hours of English audiobook recordings and corresponding transcriptions. This format is ideal for large-scale training in speech synthesis and TTS tasks.

---

### Source Data

- **Original Dataset**: [LibriTTS](https://openslr.org/60)
- **License**: The original LibriTTS dataset is licensed under the [Creative Commons Attribution 4.0 International License](https://creativecommons.org/licenses/by/4.0/). This derived dataset retains the same license.

### Modifications

- **Data Format**: The data has been restructured into larger parquet files to enhance I/O efficiency, reducing load times for distributed training environments.
- **Storage Optimization**: This derived dataset improves upon the storage requirements and retrieval efficiency, leveraging the parquet format's compression capabilities.

### Dataset Structure

- **File Format**: Parquet files.
- **Sampling Rate**: 24 kHz (same as LibriTTS).
- **Speaker Details**: Over 2,400 unique speakers with balanced representation of male and female voices, retained from LibriTTS.

### Attribution

This dataset is based on work by [LibriTTS](https://openslr.org/60), with modifications for I/O efficiency by converting to parquet file format. Please cite the original LibriTTS dataset in any publications or projects.