Hugging Face's logo

Datasets:
Hannie0813
/
NVSpeech170k

Tasks:
Text-to-Speech
Automatic Speech Recognition
Modalities:
Audio
Text
Formats:
webdataset
Languages:
Chinese
Size:
100K - 1M
Libraries:
Datasets
WebDataset
License:
Dataset card Data Studio Files Community
1
NVSpeech170k / README.md
Hannie0813's picture
Hannie0813
Upload README.md with huggingface_hub
17b36d6 verified
preview code
|
raw
history blame contribute delete
2.38 kB
metadata
license: cc-by-nc-sa-4.0
task_categories:
  - text-to-speech
  - automatic-speech-recognition
language:
  - zh

NVSpeech Dataset

Overview

The NVSpeech dataset provides extensive annotations of paralinguistic vocalizations for Mandarin Chinese speech, aimed at enhancing the capabilities of automatic speech recognition (ASR) and text-to-speech (TTS) systems. The dataset features explicit word-level annotations for 18 categories of paralinguistic vocalizations, including non-verbal sounds like laughter and breathing, as well as lexicalized interjections like "uhm" and "oh."

Dataset Description

  • NVSpeech: An automatically annotated larger subset consisting of 174,179 utterances (573.4 hours of speech). Annotations in this set are generated by a state-of-the-art paralinguistic-aware ASR model, ensuring scalability and diversity for robust model training.

Annotation Categories

The NVSpeech dataset includes annotations for the following paralinguistic vocalization categories:

  • [Breathing]
  • [Laughter]
  • [Cough]
  • [Sigh]
  • [Confirmation-en]
  • [Question-en]
  • [Question-ah]
  • [Question-oh]
  • [Surprise-ah]
  • [Surprise-oh]
  • [Dissatisfaction-hnn]
  • [Uhm]
  • [Shh]
  • [Crying]
  • [Surprise-wa]
  • [Surprise-yo]
  • [Question-ei]
  • [Question-yi]

Usage 

    from datasets import load_dataset
    dataset = load_dataset("Hannie0813/NVSpeech170k")

Intended Use

NVSpeech is designed to facilitate:

  • Training and evaluation of paralinguistic-aware speech recognition models.
  • Development of expressive and controllable TTS systems that can accurately synthesize human-like speech with inline paralinguistic cues.

Tasks

  • Automatic Speech Recognition (ASR)
  • Text-to-Speech (TTS) Synthesis
  • Paralinguistic Tagging

Languages

  • Mandarin Chinese

Dataset Structure

  • Format: Audio (WAV format) paired with text annotations including inline paralinguistic tokens.
  • Size: 174,179 automatically annotated utterances, totaling over 573 hours.

License

NVSpeech dataset is available for research use under the Creative Commons Attribution-NonCommercial-ShareAlike (CC BY-NC-SA) license.

Citation

If you use NVSpeech in your research, please cite:

Contact

For further questions, please visit the project webpage or contact the authors through the provided channels.