{"url":"/dataset/speechbrown","name":"Speech Brown","full_name":null,"description_markdown":"[![arXiv](https://img.shields.io/badge/arXiv-Paper-<COLOR>.svg)](https://arxiv.org/abs/2412.13071) [![GitHub](https://img.shields.io/badge/GitHub-Code-181717?logo=github)](https://github.com/language-modeling-lab/CLASP)\r\n\r\n## Dataset Summary\r\n\r\n**Speech Brown** is a comprehensive, synthetic, and diverse paired speech-text dataset in 15 categories, covering a wide range of topics from fiction to religion. This dataset consists of over 55,000 sentence-level samples.  \r\n\r\nTo train the [CLASP](https://huggingface.co/llm-lab/CLASP) model, we created this dataset based on the Brown Corpus. The synthetic speech was generated using the [NVIDIA Tacotron 2](https://pytorch.org/hub/nvidia_deeplearningexamples_tacotron2/) text-to-speech model.  \r\n\r\nFor more information about our proposed model, please refer to this [paper](https://arxiv.org/abs/2412.13071). The dataset generation pipeline, along with code and usage instructions, is available on this [GitHub page](https://github.com/language-modeling-lab/CLASP).  \r\n\r\n\r\n## Dataset Statistics\r\n1. Total size: Approximately 30 GB.  \r\n2. Number of samples: 55,173 pairs of speech and text.  \r\n3. Average tokens per sample: 19.00.  \r\n4. Maximum tokens in a sample: 48.  \r\n5. Average characters per sample: 96.72.\r\n6. Number of unique tokens: 50,667\r\n7. Categories: 15 categories consist of `adventure`, `belles_lettres`, `editorial`, `fiction`, `government`, `hobbies`, `humor`, `learned`, `lore`, `mystery`, `news`, `religion`, `reviews`, `romance`, `science_fiction`.  \r\n\r\n## Dataset Structure\r\nTo ensure ease of use, the dataset is partitioned into 10 parts. Each part can be used independently if it meets the requirements of your task and model.  \r\n\r\n### Metadata Files\r\n1. **global_metadata**: A JSON file containing metadata for all 55,173 samples.  \r\n2. **localized_metadata**: A JSON file containing metadata for all samples, categorized into the 10 dataset partitions.  \r\n\r\n### Metadata Fields\r\n1. **id**: The unique identifier for the sample.  \r\n2. **audio_file_path**: The file path for the audio in the dataset.  \r\n3. **category**: The category of the sample's text.  \r\n4. **text**: The corresponding text of the audio file.\r\n\r\n## Usage Instructions\r\n\r\nTo use this dataset, download the parts and metadata files as follows:\r\n\r\n#### Option 1: Manual Download\r\nVisit the [dataset repository](https://huggingface.co/datasets/llm-lab/SpeechBrown/tree/main) and download all `dataset_partX.zip` files and the `global_metadata.json` file.\r\n\r\n#### Option 2: Programmatic Download\r\nUse the `huggingface_hub` library to download the files programmatically:\r\n\r\n```python\r\nfrom huggingface_hub import hf_hub_download\r\nfrom zipfile import ZipFile\r\nimport os\r\nimport json\r\n\r\n# Download dataset parts\r\nzip_file_path1 = hf_hub_download(repo_id=\"llm-lab/SpeechBrown\", filename=\"dataset_part1.zip\", repo_type=\"dataset\")\r\nzip_file_path2 = hf_hub_download(repo_id=\"llm-lab/SpeechBrown\", filename=\"dataset_part2.zip\", repo_type=\"dataset\")\r\n# Download other parts...\r\n\r\n# Download metadata\r\nmetadata_file_path = hf_hub_download(repo_id=\"llm-lab/SpeechBrown\", filename=\"global_metadata.json\", repo_type=\"dataset\")\r\n\r\nfor i in range(1, 11):\r\n    with ZipFile(f'dataset_part{i}.zip', 'r') as zip_ref:\r\n        zip_ref.extractall(f'dataset_part{i}')\r\n    os.remove(f'dataset_part{i}.zip')\r\n\r\nwith open('global_metadata.json', 'r') as f:\r\n    metadata = json.load(f)\r\nmetadata.keys()\r\n```\r\n\r\n## Citations\r\nIf you find our paper, code, data, or models useful, please cite the paper:  \r\n```\r\n@misc{abootorabi2024claspcontrastivelanguagespeechpretraining,\r\n      title={CLASP: Contrastive Language-Speech Pretraining for Multilingual Multimodal Information Retrieval}, \r\n      author={Mohammad Mahdi Abootorabi and Ehsaneddin Asgari},\r\n      year={2024},\r\n      eprint={2412.13071},\r\n      archivePrefix={arXiv},\r\n      primaryClass={cs.CL},\r\n      url={https://arxiv.org/abs/2412.13071}, \r\n}\r\n```\r\n\r\n## Contact\r\nIf you have questions, please email mahdi.abootorabi2@gmail.com or asgari@berkeley.edu.","description_withheld":null,"homepage":"https://huggingface.co/datasets/llm-lab/SpeechBrown","introduced_date":"2024-12-17","introduced_date_note":null,"introduced_by":{"paper":"/paper/clasp-contrastive-language-speech-pretraining","title":"CLASP: Contrastive Language-Speech Pretraining for Multilingual Multimodal Information Retrieval","first_author":"Mohammad Mahdi Abootorabi","url":null},"license":{"name":"MIT","url":"https://choosealicense.com/licenses/mit/"},"modalities":[{"name":"Texts","url":"/datasets/modality/texts"},{"name":"Speech","url":"/datasets/modality/speech"}],"tasks":[{"name":"Automatic Speech Recognition (ASR)","url":"/task/automatic-speech-recognition","datasets_with_task":"/datasets/task/automatic-speech-recognition"},{"name":"Semantic Similarity","url":"/task/semantic-similarity","datasets_with_task":"/datasets/task/semantic-similarity"},{"name":"Semantic Retrieval","url":"/task/semantic-retrieval","datasets_with_task":"/datasets/task/semantic-retrieval"}],"languages":[{"name":"English","url":"/datasets/language/english"}],"variants":["Speech Brown"],"data_loaders":[],"num_papers_in_archive":1,"source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28"},"benchmarks":[],"papers_with_a_benchmark_row":[],"syntology_totals":{"read_at":"2026-09-24T18:15:14+00:00","papers_with_samples":0,"samples_harvested":0,"samples_ran":0,"samples_unverified":0,"pointer_only_for_licence":0,"papers_with_no_sample_that_ran":0,"note":"the per-paper counts above, summed; not a rate"},"papers_note":"The archive never published its papers-using-dataset list; these are papers with a leaderboard row on this dataset's benchmarks."}