README.md

metadata

dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: sentence
      dtype: string
    - name: duration
      dtype: float64
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train*
      - split: dev
        path: data/dev.*
      - split: dev_cv
        path: data/dev_cv*
      - split: dev_mls
        path: data/dev_mls*
      - split: dev_parl
        path: data/dev_parl*
      - split: dev_oslr
        path: data/dev_oslr*
      - split: dev_vp
        path: data/dev_vp*
      - split: test_cv
        path: data/test_cv*
      - split: test_mls
        path: data/test_mls*
      - split: test_parl
        path: data/test_parl*
      - split: test_oslr
        path: data/test_oslr*
      - split: test_vp
        path: data/test_vp*

Composite dataset for Spanish made from public available data

This dataset is composed of the following public available data:

Train split:

The train split is composed of the following datasets combined:

• mozilla-foundation/common_voice_18_0/es: "validated" split removing "test_cv" and "dev_cv" split's sentences. (validated split contains official train + dev + test splits and more unique data)
• openslr: a train split made from the SLR(39,61,67,71,72,73,74,75,108) subsets, this split has been cleaned from acronyms, numbers and sentences that are repeated in the following "test_oslr" and "dev_oslr" splits.
• mls/es: the "train" split from the spanish dataset of Multilingual Librispeech.
• facebook/voxpopuli/es: the "train" split from the spanish voxpopuli dataset, cleaned from acronyms and numeric characters.
• gttsehu/basque_parliament_1/es: the official "train_clean" split. The Total hours and sentences is slightly smaller because some of the sentences were removed due to be repeated in some of the test and dev splits.

  Split tag	Source	Hours	Sentences
  -	common_voice_18_0	538.82 h	378560
  -	openslr	45.58 h	24460
  -	mls	922.47 h	221855
  -	voxpopuli	142.96 h	48667
  -	basque_parliament_1	949.27 h	469937
  train	Total	2596.32 h	1142586

Test splits:

Those test splits are separated, and it is recommended to not evaluate them together in a single split:

• mozilla-foundation/common_voice_18_0/es: the official "test" split.
• openslr: a test split made from the SLR(39,61,67,71,72,73,74,75,108) subsets, this split has been cleaned from acronyms, numbers and sentences.
• mls/es: the "test" split from the spanish dataset of Multilingual Librispeech.
• facebook/voxpopuli/es: the "test" split from the spanish voxpopuli dataset, cleaned from acronyms and numeric characters.
• gttsehu/basque_parliament_1/es: the official "test" split.

  Split tag	Source	Hours	Sentences
  test_cv	common_voice_18_0	26.84 h	15872
  test_oslr	openslr	7.03 h	4107
  test_mls	mls	10 h	2385
  test_vp	voxpopuli	4.64 h	1446
  test_parl	basque_parliament_1	6.56 h	3450
  	Total	55.07 h	27260

Dev splits:

There is a dev split composed by 5 dev subsplits that are also independently accesible. It is recommended to use the combined "dev" split for development tasks since it is accurately balanced in number of hours (~5h each, a total of ~25h).

• mozilla-foundation/common_voice_18_0/es: a small dev split made from the official "dev" split.
• openslr: a small dev split made from the SLR(39,61,67,71,72,73,74,75,108) subsets, this split has been cleaned from acronyms, numbers and sentences.
• mls/es: a small "dev" split from the original "dev" split from spanish dataset of Multilingual Librispeech.
• facebook/voxpopuli/es: the original "dev" split from the spanish voxpopuli dataset, cleaned from acronyms and numeric characters.
• gttsehu/basque_parliament_1/es: the official "dev" split.

  Split tag	Source	Hours	Sentences
  dev_cv	common_voice_18_0	5.03 h	3000
  dev_oslr	openslr	5.13 h	3063
  dev_mls	mls	5.09 h	1223
  dev_vp	voxpopuli	4.89 h	1564
  dev_parl	basque_parliament_1	4.81 h	2567
  dev	Total	24.95 h	11417

Funding:

This project with reference 2022/TL22/00215335 has been parcially funded by the Ministerio de Transformación Digital and by the Plan de Recuperación, Transformación y Resiliencia – Funded by the European Union – NextGenerationEU ILENIA and by the project IkerGaitu funded by the Basque Government.