Speech Recognition Alignment Dataset

This dataset is a variation of several widely-used ASR datasets, encompassing Librispeech, MuST-C, TED-LIUM, VoxPopuli, Common Voice, and GigaSpeech. The difference is this dataset includes:

• Precise alignment between audio and text.
• Text that has been punctuated and made case-sensitive.
• Identification of named entities in the text.

Usage

First, install the latest version of the 🤗 Datasets package:

pip install --upgrade pip
pip install --upgrade datasets[audio]

The dataset can be downloaded and pre-processed on disk using the load_dataset function:

from datasets import load_dataset

# Available dataset: 'libris','mustc','tedlium','voxpopuli','commonvoice','gigaspeech'
dataset = load_dataset("nguyenvulebinh/asr-alignment", "libris")

# take the first sample of the validation set
sample = dataset["train"][0]

It can also be streamed directly from the Hub using Datasets' streaming mode. Loading a dataset in streaming mode loads individual samples of the dataset at a time, rather than downloading the entire dataset to disk:

from datasets import load_dataset

dataset = load_dataset("nguyenvulebinh/asr-alignment", "libris", streaming=True)
# take the first sample of the validation set
sample = next(iter(dataset["train"]))

Citation

If you use this data, please consider citing the ICASSP 2024 Paper: SYNTHETIC CONVERSATIONS IMPROVE MULTI-TALKER ASR:

@INPROCEEDINGS{synthetic-multi-asr-nguyen,
  author={Nguyen, Thai-Binh and Waibel, Alexander},
  booktitle={ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, 
  title={SYNTHETIC CONVERSATIONS IMPROVE MULTI-TALKER ASR}, 
  year={2024},
  volume={},
  number={},
}

License

This dataset is licensed in accordance with the terms of the original dataset.