metadata

license: mit
language:
  - de
size_categories:
  - 10K<n<100K
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
dataset_info:
  features:
    - name: speaker
      dtype: string
    - name: book
      dtype: string
    - name: audio
      dtype: audio
    - name: text
      dtype: string
  splits:
    - name: train
      num_bytes: 74381179391.329
      num_examples: 96331
  download_size: 77346868399
  dataset_size: 74381179391.329

HUI-Audio-Corpus-German Dataset

Overview

The HUI-Audio-Corpus-German is a high-quality Text-To-Speech (TTS) dataset developed by researchers at the Institute of Information Systems (IISYS). This dataset is designed to facilitate the development and training of TTS applications, particularly in the German language. The associated research paper can be found here.

Dataset Contents

The dataset comprises recordings from multiple speakers, with the five most prominent speakers listed individually and the remaining grouped under "Others." Each audio file is sampled at a rate of 44.1 kHz. For each speaker, there is a "Full" dataset and a "Clean" variant, the latter offering even higher audio quality. Detailed statistics for each subset are provided below.

Speakers and Data Availability

Bernd Ungerer
- Full: 97 hours | 23 GB
- Clean: 92 hours | 22 GB
Hokuspokus
- Full: 43 hours | 10 GB
- Clean: 27 hours | 7 GB
Friedrich
- Full: 32 hours | 7 GB
- Clean: 21 hours | 5 GB
Karlsson
- Full: 30 hours | 7 GB
- Clean: 29 hours | 7 GB
Eva K
- Full: 29 hours | 7 GB
- Clean: 22 hours | 5 GB
Others
- Full: 96 hours | 23 GB
- Clean: 64 hours | 15 GB

Download Links

Full Dataset:

Clean Dataset:

Disclaimer

This dataset was obtained from university research. I do not accept any legal responsibility for its use. If there are any concerns or requests for removal, please contact me at [email protected], and I will address the issue promptly.