Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,42 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
pretty_name: Speech Brown
|
| 6 |
+
size_categories:
|
| 7 |
+
- 10K<n<100K
|
| 8 |
+
task_categories:
|
| 9 |
+
- text-to-speech
|
| 10 |
+
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
## Dataset Summary
|
| 14 |
+
|
| 15 |
+
**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.
|
| 16 |
+
|
| 17 |
+
To 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.
|
| 18 |
+
|
| 19 |
+
For 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).
|
| 20 |
+
|
| 21 |
+
|
| 22 |
+
|
| 23 |
+
## Dataset Statistics
|
| 24 |
+
1. Total size: Approximately 30 GB.
|
| 25 |
+
2. Number of samples: 55,173 pairs of speech and text.
|
| 26 |
+
3. Average words per sample: 17.78.
|
| 27 |
+
4. Maximum words in a sample: 48.
|
| 28 |
+
5. Average characters per sample: 96.72.
|
| 29 |
+
6. Categories: `adventure`, `belles_lettres`, `editorial`, `fiction`, `government`, `hobbies`, `humor`, `learned`, `lore`, `mystery`, `news`, `religion`, `reviews`, `romance`, `science_fiction`.
|
| 30 |
+
|
| 31 |
+
## Dataset Structure
|
| 32 |
+
To 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.
|
| 33 |
+
|
| 34 |
+
### Metadata Files:
|
| 35 |
+
1. **global_metadata**: A JSON file containing metadata for all 55,173 samples.
|
| 36 |
+
2. **localized_metadata**: A JSON file containing metadata for all samples, categorized into the 10 dataset partitions.
|
| 37 |
+
|
| 38 |
+
### Metadata Fields:
|
| 39 |
+
1. **id**: The unique identifier for the sample.
|
| 40 |
+
2. **audio_file_path**: The file path for the audio in the dataset.
|
| 41 |
+
3. **category**: The category of the sample's text.
|
| 42 |
+
4. **text**: The corresponding text of the audio file.
|