camoes_SI

Dataset Description

The camoes_SI dataset is a curated combination of two European Portuguese sociolinguistic corpora --- Fala Bracarense and Português Fundamental --- merged into a unified test-only dataset for evaluating Automatic Speech Recognition (ASR) systems.

All audio is provided as 16 kHz PCM waveforms, accompanied by speaker metadata and reference transcripts.

This dataset corresponds to the Sociolinguistic Interviews (SI) category of the CAMÕES benchmark, introduced in:

CAMÕES: A Comprehensive Automatic Speech Recognition Benchmark for European Portuguese

In the CAMÕES benchmark, the SI category represents:

Highly spontaneous conversational speech, recorded in various Portuguese regions and social contexts, often with poor recording conditions and highly accented speech making this the most challening domain in this benchmark.

As mention above, SI is the most difficult domain in the benchmark and a strong indicator of model robustness in real conversational scenarios.

---

Source Components

The dataset here on Hugging Face is the union of two corpora, both referenced in the CAMÕES benchmark:

Fala Bracarense

• Duration: 6.1 hours
• Speakers: 9
• Age range: 15--92
• Gender: 45% male | 55% female
• Location: Braga, Portugal
• Collection period: 2009--2014

Português Fundamental

• Duration: 4.2 hours
• Speakers: 169
• Age range: 17--69
• Gender: 44% male | 56% female
• Collection period: 1970s

---

Data Fields

Column Name	Type	Description
audio	Audio	16 kHz PCM waveform of the utterance
age	string	Speaker-reported age
gender	string	Speaker gender
speaker_id	string	Unique speaker identifier
hypothesis	string	ASR hypothesis transcript (if available)
reference	string	Ground-truth transcript
wrd	string	Word count or related metric
wer	string	Word Error Rate (if available)
dataset	string	Source dataset identifier
ID	string	Unique utterance ID
ncount	string	Additional metadata field
sex	string	Speaker sex / alternative gender field

---

Example Usage

from datasets import load_dataset, Audio

# Load full test split
ds = load_dataset("inesc-id/camoes_SI", split="test")

# Filter for FBracarense or PTFundamental or keep both and skip this line
fb = ds.filter(lambda x: x["dataset"] == "FBracarense")

---

License

This dataset is released under the CC BY-NC 4.0 license.
You may use, modify, and redistribute the dataset for non-commercial research purposes.
Commercial use is not permitted.

---

References

Fala Bracarense

@misc{FalaBracarense2009,
  author       = {{Centro de Estudos Humanísticos, Universidade do Minho}},
  title        = {Perfil Sociolinguístico da Fala Bracarense},
  year         = {2009},
  howpublished = {\url{https://sites.google.com/site/projectofalabracarense/}},
  note         = {Accessed: 2025-05-21}
}

Português Fundamental

@misc{PTFundamental2014,
  author       = {{Centro de Linguística, Universidade de Lisboa}},
  title        = {Português Fundamental},
  year         = {2014},
  howpublished = {\url{https://www.islrn.org/resources/812-337-422-842-3/}},
  note         = {Accessed: 2025-05-21}
}

CAMÕES Benchmark

@inproceedings{2025CAMOES,
  title={CAMÕES: A Comprehensive Automatic Speech Recognition Benchmark for European Portuguese},
  author={Carvalho, Carlos and Teixeira, Francisco and Botelho, Catarina and Pompili, Anna and Solera-Ureña, Rubén and Paulo, Sérgio and Julião, Mariana and Rolland, Thomas and Mendonça, John and Pereira, Diogo and Trancoso, Isabel and Abad, Alberto},
  booktitle={IEEE Automatic Speech Recognition and Understanding Workshop (ASRU)},
  year={2025}
}