meetween/mumospee_gigaspeech

Dataset Summary

This dataset is a modified version of the gigaspeech corpus, converted into parquet format to facilitate optimized I/O operations in high-performance and distributed computing environments. GigaSpeech is an evolving, multi-domain English speech recognition corpus with 10,000 hours of high quality labeled audio suitable for supervised training. The transcribed audio data is collected from audiobooks, podcasts and YouTube, covering both read and spontaneous speaking styles, and a variety of topics, such as arts, science, sports, etc.

---

Source Data

• Original Dataset: gigaspeech
• License: This derived dataset is shared under the same license, with modifications only to format for efficiency.

Modifications

• Data Format: Converted to parquet format to enhance I/O performance for distributed training, reducing latency during data loading and retrieval.
• Efficiency Optimization: Restructured for reduced storage footprint and faster I/O on high-performance clusters by leveraging parquet’s efficient compression and columnar storage.

Dataset Structure

• File Format: Parquet files.
• Languages: English
• Audio Sampling Rate: Matches original dataset specifications for high-fidelity speech data.

mumospee_gigaspeech/
├── gigaspeech-parquet/      # audio + full source fields
│   ├── gigaspeech.xl.train-00000-of-00061.parquet ... train-00060-of-00061.parquet
│   ├── test.parquet
│   └── validation.parquet
└── metadata/                # lightweight metadata index, one parquet per split
    ├── train.parquet
    ├── test.parquet
    └── validation.parquet

The metadata/ parquet files index the audio segments (the url column records the source audio shard for each segment) and contain the following columns:

column	type	description
path	string	{segment_id}.wav
url	string	URL of the audio parquet shard containing this segment
type	string	always audio
duration	float64	segment duration in seconds (end_time - begin_time)
language	string	always en
transcript	string	transformed transcription (truecased, punctuation restored)
tag	string	always GigaSpeech
split	string	train / test / validation
license	string	always apache-2.0

Total segments: 6,042,262 (train 6,016,616 · test 19,931 · validation 5,715).

Usage

This dataset is ideal for use in large-scale speech-to-text translation tasks, especially in distributed and high-performance computing environments. The parquet format enhances usability by minimizing I/O overhead, making it well-suited for high-throughput training.

Attribution

This dataset is based on the original gigaspeech dataset, with modifications for I/O optimization by converting to parquet format. Please cite the original GigaSpeech dataset in any publications or projects using this dataset.

Citation Information

Please cite this paper if you find this work useful:

@inproceedings{GigaSpeech2021,
  title={GigaSpeech: An Evolving, Multi-domain ASR Corpus with 10,000 Hours of Transcribed Audio},
  booktitle={Proc. Interspeech 2021},
  year=2021,
  author={Guoguo Chen, Shuzhou Chai, Guanbo Wang, Jiayu Du, Wei-Qiang Zhang, Chao Weng, Dan Su, Daniel Povey, Jan Trmal, Junbo Zhang, Mingjie Jin, Sanjeev Khudanpur, Shinji Watanabe, Shuaijiang Zhao, Wei Zou, Xiangang Li, Xuchen Yao, Yongqing Wang, Yujun Wang, Zhao You, Zhiyong Yan}
}

Changelog

2026-06-29

• Removed pure non-speech segments (transcript = a single <MUSIC>/<NOISE>/<SIL>/<OTHER> marker, no spoken words) from the test and validation splits. This also removed the 5 zero-duration segments. New total: 6,042,262.
• Consolidated the 61 per-shard train metadata parquets into a single metadata/train.parquet.

2026-06-27

• Regenerated the metadata/ index as parquet (one file per audio shard), replacing the previous CSV metadata.