Datasets:

Modalities:
Audio
Text
Formats:
parquet
Languages:
German
Size:
< 1K
Libraries:
Datasets
pandas
License:
swiss-ner / README.md
kenfus's picture
Upload dataset
352878e verified
metadata
dataset_info:
  features:
    - name: audio
      dtype:
        audio:
          sampling_rate: 16000
    - name: sentence
      dtype: string
    - name: named_entities
      sequence: string
    - name: entity_types
      sequence: string
    - name: entity_languages
      sequence: string
  splits:
    - name: test
      num_bytes: 17175733
      num_examples: 172
  download_size: 12084820
  dataset_size: 17175733
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test-*
license: cc-by-4.0
language:
  - de
size_categories:
  - n<1K

SwissNER-Spoken is a curated collection of 173 short, spoken-style German sentences designed to evaluate Named-Entity Recognition (NER) and Automatic Speech Recognition (ASR) systems on Swiss-specific proper nouns.

Key features

Focus on Switzerland – Every sentence contains up to three named entities that appear in everyday Swiss contexts: cities, villages, cantons, companies, mountains, lakes, rivers, landmarks, organizations, events and well-known personalities.

Balanced regional coverage – All 26 Swiss cantons are represented, with entities drawn from German-, French-, Italian- and Romansh-speaking areas.

Multilingual entity labels – For each entity, the dataset records the language of the name (de, fr, it, rm or en) so models can test language-aware recognition.

Compact schema – Four CSV columns: 1. text – the sentence in German. 2. named_entities – comma-separated list of the entities in the sentence. 3. entity_types – aligned list of coarse entity classes (e.g. City, Company, Mountain). 4. entity_languages – aligned list of language tags for each entity name.

ASR-oriented style – Sentences are intentionally short, natural, and pronunciation- friendly, making the corpus ideal for measuring how well speech or text models handle Swiss proper nouns in real-world utterances.

Typical row

"Nestlé hat seinen Sitz in Vevey im Kanton Waadt.", "Nestlé, Vevey, Waadt", "Company, City, Canton", "fr, fr, de"

Use cases

  • Benchmarking NER models on Swiss entities
  • Stress-testing ASR/voice pipelines for pronunciation and transcription accuracy
  • Data augmentation or few-shot prompts for multilingual Swiss NLP tasks
  • Educational demos for Swiss geography, culture and corporate landscape

License

CC-BY-4.0 – attribution required; no additional restrictions.