Dataset Card for PELCRA benchmark corpora

Name: pl-asr-pelcra-for-bigos
Creator: PELCRA
License: https://choosealicense.com/licenses/cc-by-nc-nd-4.0/

Dataset Summary

This repository contains subsets of the SpokesMix (SpokesMix, SpokesBiz and Diabiz) corpora processed in the format of the BIGOS (Benchmark Intended Grouping of Open Speech) format. The main contribution of these corpora are a) spontaneous and conversional speech and b) phone-based customer interactions. The inclusion of the subsets in the BIGOS corpora, will hopefully result in the most comprehensive publicly available evaluation of Polish ASR systems in terms of number of speakers, devices and acoustic conditions. Further contributions to the benchmark datasets from the community are highly welcomed.

Supported Tasks and Leaderboards

The PELCRA test sets are intented for benchmarking Polish ASR systems under the 2024 PolEval challenge.

Languages

Polish

Dataset Structure

The datasets consist of audio recordings in the WAV format with corresponding metadata.
The audio and metadata can be used in a raw format (TSV) or via the hugging face datasets library.
References for the test split will only become available after the completion of the 2024 PolEval challenge.

Data Instances

Data Fields

Available fields:

audioname - file identifier
split - test, validation or train split
dataset - source dataset identifier
audio - HF dataset object with binary representation of audio file
ref_orig - original transcription of audio file
samplingrate_orig - sampling rate of the original recording
sampling_rate - sampling rate of recording in the release
audio_duration_samples - duration of recordings in samples
audio_duration_seconds - duration of recordings in seconds
audiopath_bigos - relative filepath to audio file extracted from tar.gz archive
audiopath_local - absolute filepath to audio file extracted with the build script
speaker_gender - gender (sex) of the speaker extracted from the source meta-data (N/A if not available)
speaker_age - age group of the speaker (in CommonVoice format) extracted from the source (N/A if not available)
utt_length_words - length of the utterance in words
utt_length_chars - length of the utterance in characters
speech_rate_words - ratio of words to recording duration.
speech_rate_chars - ratio of characters to recording duration.

Data Splits

Train split contains recordings intendend for training. Validation split contains recordings for validation during training procedure. Test split contains recordings intended for evaluation only. References for test split are not available until the completion of 2024 PolEval challenge.

Subset	train	validation	test
ul-diabiz_poleval-22	7 719	284	947
ul-spokes_biz_bio-23	45 488	3 910	5 519
ul-spokes_biz_int-23	726	82	301
ul-spokes_biz_luz-23	32 539	5 528	3 899
ul-spokes_biz_pod-23	19 511	2 260	1 036
ul-spokes_biz_pres-23	14 165	1 780	1 229
ul-spokes_biz_vc-23	36 897	2 819	5 556
ul-spokes_biz_vc2-23	20 187	2 384	3 231
ul-spokes_biz_wyw-23	9 711	1 501	145
ul-spokes_mix_emo-18	20 186	3 147	996
ul-spokes_mix_luz-18	14 840	4 324	1 755
ul-spokes_mix_parl-18	7 181	513	962

Dataset Creation

Curation Rationale

The dataset was curated in order to enable convenient use of training, validation and test corpora across all publically available ASR speech datasets for Polish. Thanks to similar format of PELCRA and BIGOS corpora and convenient access via hugging face platform ASR practioners are able to leverage vast amount of diverse set of recordings with reduced overhead. Recordings which either lacked transcriptions or were too short to be useful for training or evaluation were removed during curation.

Source Data

Data was sourced from the below original datasets with the original authors' permission:

SpokesMix corpus
SpokesBiz corpus.
PolEval DiaBiz sample corpus.

Initial Data Collection and Normalization

Source text and audio files were extracted and encoded in a unified format.
Original dataset-specific transcription norms are preserved, including punctuation and casing.

Who are the source language producers?

The PELCRA SpokesMix, SpokesBiz and DiaBiz datasets were created by:
Piotr Pęzik, Michał Adamczyk, Małgorzata Krawentek, Paweł Wilk, Sylwia Karasińska, Angelika Peljak-Łapińska, Karolina Adamczyk, Monika Garnys, Karolina Walkusz, Anna Cichosz, Anna Kwiatkowska, Mikołaj Deckert, Paulina Rybińska, Izabela Grabarczyk, Maciej Grabski, Karol Ługowski, Michał Koźmiński, Zuzanna Deckert, Piotr Górniak, Konrad Kaczyński, Łukasz Jałowiecki and others.

If you use this dataset, please cite the data curator, as well the original authors:

@misc {pelcra_2024,
    author       = { {Piotr Pęzik, Michał Junczyk} },
    title        = { pl-asr-pelcra-for-bigos (Revision 4205ec7) },
    year         = 2024,
    url          = { https://huggingface.co/datasets/pelcra/pl-asr-pelcra-for-bigos },
    doi          = { 10.57967/hf/2387 },
    publisher    = { Hugging Face }
}

@misc{pęzik2023spokesbizopencorpus,
      title={SpokesBiz -- an Open Corpus of Conversational Polish}, 
      author={Piotr Pęzik and Sylwia Karasińska and Anna Cichosz and Łukasz Jałowiecki and Konrad Kaczyński and Małgorzata Krawentek and Karolina Walkusz and Paweł Wilk and Mariusz Kleć and Krzysztof Szklanny and Szymon Marszałkowski},
      year={2023},
      eprint={2312.12364},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2312.12364}, 
}

@InProceedings{PĘZIK18.888,
  author = {Piotr Pęzik},
  title = "{Increasing the Accessibility of Time-Aligned Speech Corpora with Spokes Mix}",
  booktitle = {Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)},
  year = {2018},
  month = {May 7-12, 2018},
  address = {Miyazaki, Japan},
  editor = {Nicoletta Calzolari (Conference chair) and Khalid Choukri and Christopher Cieri and Thierry Declerck and Sara Goggi and Koiti Hasida and Hitoshi Isahara and Bente Maegaard and Joseph Mariani and Hélène Mazo and Asuncion Moreno and Jan Odijk and Stelios Piperidis and Takenobu Tokunaga},
  publisher = {European Language Resources Association (ELRA)},
  isbn = {979-10-95546-00-9},
  language = {english}
  }

@inproceedings{Pezik2022,
   abstract = {This paper introduces DiaBiz, a large, annotated, multimodal corpus of Polish telephone conversations conducted in varied business settings, comprising 4036 call centre interactions from nine different domains, i.e. banking, energy services, telecommunications, insurance, medical care, debt collection, tourism, retail and car rental. The corpus was developed to boost the development of third-party speech recognition engines, dialog systems and conversational intelligence tools for Polish. Its current size amounts to nearly 410 hours of recordings and over 3 million words of transcribed speech. We present the structure of the corpus, data collection and transcription procedures, challenges of punctuating and truecasing speech transcripts, dialog structure annotation and discuss some of the ecological validity considerations involved in the development of such resources.},
   author = {Piotr Pezik and Gosia Krawentek and Sylwia Karasińska and Paweł Wilk and Paulina Rybińska and Anna Cichosz and Angelika Peljak-Łapińska and Mikołaj Deckert and Michał Adamczyk},
   city = {Marseille, France},
   editor = {Nicoletta Calzolari and Frédéric Béchet and Philippe Blache and Khalid Choukri and Christopher Cieri and Thierry Declerck and Sara Goggi and Hitoshi Isahara and Bente Maegaard and Joseph Mariani and Hélène Mazo and Jan Odijk and Stelios Piperidis},
   booktitle = {Proceedings of the Thirteenth Language Resources and Evaluation Conference},
   month = {6},
   pages = {723-726},
   publisher = {European Language Resources Association},
   title = {DiaBiz – an Annotated Corpus of Polish Call Center Dialogs},
   url = {https://aclanthology.org/2022.lrec-1.76},
   year = {2022},
}

The curation and inclusion of subsets of these corpora into the BIGOS format was done by Michał Junczyk from Adam Mickiewicz University.

Annotations

Annotation process

Current release contains original, manually performed transcriptions.

Who are the annotators?

Participants of the original projects.

Personal and Sensitive Information

This corpus does not contain PII or Sensitive Information. All IDs of speakers are anonymized.

Considerations for Using the Data

Social Impact of Dataset

To be updated.

Discussion of Biases

To be updated.

Other Known Limitations

Some metadata avaiable in the original releases is not available in the initial HF release e.g. speaker age, gender etc.

Additional Information

Dataset Curators

Original authors of the source datasets: Piotr Pęzik (piotr.pezik@uni.lodz.pl) et al. please refer to source-data for details. Michał Junczyk (michal.junczyk@amu.edu.pl) - curator of PELCRA corpora in the BIGOS format.

Licensing Information

The BIGOS benchmark is available under Creative Commons By Attribution Non Commercial No Derivates

Datasets:

pelcra
/

pl-asr-pelcra-for-bigos

You need to agree to share your contact information to access this dataset