Datasets:

Modalities:
Audio
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
Dask
slue-phase-2 / README.md
sshon-asapp's picture
Update README.md
e2989c5 verified
metadata
dataset_info:
  - config_name: hvb
    features:
      - name: issue_id
        dtype: string
      - name: audio
        dtype:
          audio:
            sampling_rate: 16000
      - name: speaker_id
        dtype: string
      - name: text
        dtype: string
      - name: utt_index
        dtype: int32
      - name: channel
        dtype: int32
      - name: role
        dtype: string
      - name: start_ms
        dtype: int32
      - name: duration_ms
        dtype: int32
      - name: intent
        dtype: string
      - name: dialog_acts
        sequence: string
    splits:
      - name: train
        num_bytes: 803631533.648
        num_examples: 11344
      - name: validation
        num_bytes: 115999281.63
        num_examples: 1690
      - name: test
        num_bytes: 413280185.739
        num_examples: 6121
    download_size: 1287263357
    dataset_size: 1332911001.017
  - config_name: sqa5
    features:
      - name: question_id
        dtype: string
      - name: question_audio
        dtype:
          audio:
            sampling_rate: 16000
      - name: question_speaker_id
        dtype: string
      - name: raw_question_text
        dtype: string
      - name: normalized_question_text
        dtype: string
      - name: document_id
        dtype: string
      - name: document_audio
        dtype:
          audio:
            sampling_rate: 16000
      - name: document_speaker_id
        dtype: string
      - name: raw_document_text
        dtype: string
      - name: normalized_document_text
        dtype: string
      - name: word2time
        sequence:
          - name: word
            dtype: string
          - name: normalized_word
            dtype: string
          - name: start_second
            dtype: float64
          - name: end_second
            dtype: float64
      - name: answer_spans
        sequence:
          - name: answer
            dtype: string
          - name: start_second
            dtype: float64
          - name: end_second
            dtype: float64
    splits:
      - name: train
        num_bytes: 134775904845.04
        num_examples: 46186
      - name: validation
        num_bytes: 5686714785.843
        num_examples: 1939
      - name: test
        num_bytes: 6967375359.628
        num_examples: 2382
      - name: verified_test
        num_bytes: 1182628989
        num_examples: 408
    download_size: 118074473123
    dataset_size: 148612623979.511
  - config_name: ted
    features:
      - name: id
        dtype: string
      - name: audio
        dtype:
          audio:
            sampling_rate: 16000
      - name: speaker
        dtype: string
      - name: transcript
        dtype: string
      - name: title
        dtype: string
      - name: abstract
        dtype: string
    splits:
      - name: train
        num_bytes: 46573026086.984
        num_examples: 3384
      - name: validation
        num_bytes: 5694199931
        num_examples: 425
      - name: test
        num_bytes: 5959094411
        num_examples: 423
    download_size: 58384489268
    dataset_size: 58226320428.984
  - config_name: vp_nel
    features:
      - name: id
        dtype: string
      - name: audio
        dtype:
          audio:
            sampling_rate: 16000
      - name: speaker_id
        dtype: string
      - name: text
        dtype: string
      - name: word_timestamps
        sequence:
          - name: word
            dtype: string
          - name: start_sec
            dtype: float64
          - name: end_sec
            dtype: float64
      - name: ne_timestamps
        sequence:
          - name: ne_label
            dtype: string
          - name: start_char_idx
            dtype: int32
          - name: char_offset
            dtype: int32
          - name: start_sec
            dtype: float64
          - name: end_sec
            dtype: float64
    splits:
      - name: validation
        num_bytes: 83371882.75
        num_examples: 1750
      - name: test
        num_bytes: 85222143.142
        num_examples: 1838
    download_size: 165119242
    dataset_size: 168594025.89200002
configs:
  - config_name: hvb
    data_files:
      - split: train
        path: hvb/train-*
      - split: validation
        path: hvb/validation-*
      - split: test
        path: hvb/test-*
  - config_name: sqa5
    data_files:
      - split: train
        path: sqa5/train-*
      - split: validation
        path: sqa5/validation-*
      - split: test
        path: sqa5/test-*
      - split: verified_test
        path: sqa5/verified_test-*
  - config_name: ted
    data_files:
      - split: train
        path: ted/train-*
      - split: validation
        path: ted/validation-*
      - split: test
        path: ted/test-*
  - config_name: vp_nel
    data_files:
      - split: validation
        path: vp_nel/validation-*
      - split: test
        path: vp_nel/test-*

Dataset description

Licensing Information

SLUE-HVB

SLUE-HVB dataset contains a subset of the Gridspace-Stanford Harper Valley speech dataset and the copyright of this subset remains the same with the original license, CC-BY-4.0. See also original license notice (https://github.com/cricketclub/gridspace-stanford-harper-valley/blob/master/LICENSE)

Additionally, we provide dialog act classification annotation and it is covered with the same license as CC-BY-4.0.

SLUE-SQA-5

SLUE-SQA-5 Dataset contains question texts and answer strings (question_text, normalized_question_text, and answer_spans column in .tsv files) from these datasets, * SQuAD1.1 (for questions whose question_id starts with ‘squad-’) * Natural Questions (for questions whose question_id starts with ‘nq-’) * WebQuestions (for questions whose question_id starts with ‘wq-’) * CuratedTREC (for questions whose question_id starts with ‘trec-’) * TriviaQA (for questions whose question_id starts with ‘triviaqa-’) Additionally, we provide audio recordings (.wav files in “question” directories) of these questions.

For questions from TriviaQA (questions whose question_id starts with ‘triviaqa-’), their question texts, answer strings, and audio recordings are licensed with the same Apache License 2.0 as TriviaQA (for more detail, please refer to https://github.com/mandarjoshi90/triviaqa/blob/master/LICENSE). For questions from the other 4 datasets, their question texts, answer strings, and audio recordings are licensed with Creative Commons Attribution-ShareAlike 4.0 International license.

SLUE-SQA-5 also contains a subset of Spoken Wikipedia, including the audios placed in “document” directories and their transcripts (document_text and normalized_document_text column in .tsv files). Additionally, we provide the text-to-speech alignments (.txt files in “word2time” directories).These contents are licensed with the same Creative Commons (CC BY-SA 4.0) license as Spoken Wikipedia.

SLUE-TED

SLUE-TED Dataset contains TED Talk audios along with the associated abstracts and title, which were concatenated to create reference summaries. This corpus is licensed with the same Creative Commons (CC BY–NC–ND 4.0 International) license as TED talks. For further information, please refer to the details provided below.

============================= TED.com

We encourage you to share TED Talks, under our Creative Commons license, or ( CC BY–NC–ND 4.0 International, which means it may be shared under the conditions below: CC: means the type of license rights associated with TED Talks, or Creative Commons BY: means the requirement to include an attribution to TED as the owner of the TED Talk and include a link to the talk, but do not include any other TED branding on your website or platform, or language that may imply an endorsement. NC: means you cannot use TED Talks in any commercial context or to gain any type of revenue, payment or fee from the license sublicense, access or usage of TED Talks in an app of any kind for any advertising, or in exchange for payment of any kind, including in any ad supported content or format. ND: means that no derivative works are permitted so you cannot edit, remix, create, modify or alter the form of the TED Talks in any way. This includes using the TED Talks as the basis for another work, including dubbing, voice-overs, or other translations not authorized by TED. You may not add any more restrictions that we have placed on the TED site content, such as additional legal or technological restrictions on accessing the content.