German speech emotion recognition datasets

KuralHub lists 1 speech emotion recognition dataset for German. Each entry below gives the emotion labels, size, recording setup, licence, where to get it and how to cite it. German is also part of the benchmark. The best of 11 models reached 86.0% test accuracy (whisper-large).

Language family: Indo-European (Germanic)
ISO code: de
Spoken mainly in: Germany
Datasets listed: 1
In benchmark: Yes

Datasets at a glance

Dataset	Type	Samples	Speakers	Recording	Access	Licence
EmoDB	Audio	535	10	Acted	Open

Benchmark results

German is one of the 29 languages in the KuralHub benchmark. Each of the 11 frozen speech models was trained with a small classification head on the shared emotions and scored on a held-out test split. Over all 11 runs the mean test accuracy was 55.5% (median 53.0%). The best run reached 86.0% with whisper-large. Compare with other languages.

**EmoDB** (run ID `EmoDB`). Scores on the test split, best run highlighted.
Model	Family	Val acc.	Test acc.	Macro-F1	Weighted-F1	Logs
hubert-base-ls960	HuBERT	0.75	0.78	0.76	0.78	logs
hubert-large-ls960-ft	HuBERT	0.43	0.44	0.39	0.38	logs
wav2vec2-base	wav2vec 2.0	0.68	0.70	0.68	0.70	logs
wav2vec2-large-960h	wav2vec 2.0	0.31	0.34	0.30	0.30	logs
wav2vec2-large-lv60	wav2vec 2.0	0.30	0.26	0.09	0.11	logs
wav2vec2-xls-r-300m	wav2vec 2.0	0.33	0.35	0.26	0.27	logs
wav2vec2-xls-r-1b	wav2vec 2.0	0.33	0.30	0.28	0.24	logs
wavlm-base-plus	WavLM	0.71	0.70	0.64	0.67	logs
wavlm-large	WavLM	0.52	0.53	0.44	0.44	logs
whisper-small	Whisper	0.82	0.84	0.83	0.84	logs
whisper-largebest	Whisper	0.84	0.86	0.85	0.86	logs

EmoDB

Type: Audio
Emotion labels: Anger, Boredom, Anxiety, Happy, Sadness, Disgust, and Neutral
Size: samples - 535 speakers - 10
Speakers: 10
Recording: Acted
Access: Open
Licence: Not stated
File format: .wav
Website: Dataset Homepage
Paper: Paper Link
Download: Download

The EMODB database is a freely available German emotional speech dataset created by the Institute of Communication Science at the Technical University of Berlin, Germany. It contains 535 utterances recorded by 10 professional speakers (5 males and 5 females), representing seven emotions: anger, boredom, anxiety, happiness, sadness, disgust, and neutral. The recordings were made at a 48-kHz sampling rate and later down-sampled to 16 kHz. This dataset is widely used for emotion recognition tasks in speech processing.

Used in the KuralHub benchmark (run ID EmoDB). See the results above.

Citation

@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}
}

Source: datasets/German/EmoDB.md · Suggest an edit

Know another German speech emotion dataset, or a newer version of one listed here? Suggest it or open a pull request.

This page is generated from datasets/German/ in the KuralHub repository. If you use a dataset, cite its authors. If KuralHub helped you find it, please also cite the KuralHub paper.