metadata

dataset_info:
  features:
    - name: audio
      dtype:
        audio:
          sampling_rate: 16000
    - name: emotion
      dtype: string
    - name: label
      dtype:
        class_label:
          names:
            '0': anxiety
            '1': disgust
            '2': happiness
            '3': boredom
            '4': neutral
            '5': sadness
            '6': anger
  splits:
    - name: train
      num_bytes: 26772110
      num_examples: 304
    - name: test
      num_bytes: 20866781
      num_examples: 231
  download_size: 46818101
  dataset_size: 47638891
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
task_categories:
  - audio-classification
tags:
  - audio
  - paralinguistic
  - multiclass

EmoDB

The EmoDB is the freely available German emotional database, containing a total of 535 utterances. It comprises of seven emotions: 1) anger; 2) boredom; 3) anxiety; 4) happiness; 5) sadness; 6) disgust; and 7) neutral. The data was recorded at a 48-kHz sampling rate and then down-sampled to 16-kHz. We follow the unofficial speaker-independent train/test split from here.

Citations

@inproceedings{burkhardt2005database,
  title={A database of German emotional speech.},
  author={Burkhardt, Felix and Paeschke, Astrid and Rolfes, Miriam and Sendlmeier, Walter F and Weiss, Benjamin and others},
  booktitle={Interspeech},
  volume={5},
  pages={1517--1520},
  year={2005}
}