sTinyStories

A spoken version of TinyStories Synthesized with LJ voice using FastSpeech2. The dataset was synthesized to boost the training of Speech Language Models as detailed in the paper "Slamming: Training a Speech Language Model on One GPU in a Day". It was first suggested by Cuervo et. al 2024.

We refer you to the SlamKit codebase to see how you can train a SpeechLM with this dataset.

Usage

from datasets import load_dataset
dataset = load_dataset("slprl/sTinyStories")

sTinyStories is big, you can download parital sets using the streaming=True variable.

dataset = load_dataset("slprl/sTinyStories", split="train", streaming=True)
head = dataset.take(5)  # takes the first 5 rows

for row in head:
    print(row)

You can read more about handling large datasets using the streaming variable here.

Another possibility to download a partial set using hf_hub_download:

from huggingface_hub import list_repo_files, hf_hub_download
from datasets import Dataset

files = list_repo_files("slprl/sTinyStories", repo_type="dataset")
parquet_files = [f for f in files if f.endswith('parquet')]
sample = hf_hub_download("slprl/sTinyStories", parquet_files[0], repo_type="dataset")

ds = Dataset.from_parquet(sample)

Data Fields

The data has several fields:

• audio: The synthesized audio.
• text: The text of the prompt audio.
• aligned_text: The text of the prompt audio, aligned to the time in the audio: [("word", "start_time", "end_time") ... ]
• split_sentence: boolean - some of the stories were generated while splitting the sentences to provide better audio quality and seperation between setnences. Due to compute budget, we generated only some of the texts with this technique.

Citation

If you use this version of the dataset please cite our work:

@misc{maimon2025slamming,
      title={Slamming: Training a Speech Language Model on One GPU in a Day}, 
      author={Gallil Maimon and Avishai Elmakies and Yossi Adi},
      year={2025},
      eprint={2502.15814},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2502.15814}, 
}