{"success":true,"data":{"author":"KBLab, National Library of Sweden","bucketAccessType":"public","bucketName":"KBLab/rixvox-v2","bucketPrefix":"","category":"Speech & Audio","cloudProvider":"huggingface","createdAt":"2026-11-14T14:15:18.210Z","description":"RixVox-v2 is a Swedish parliamentary speech dataset spanning nearly 23,000 hours of speech, built by KBLab at the National Library of Sweden by matching and force-aligning speeches in the Riksdag's parliamentary protocols to media recordings of the debates. Each observation carries metadata about the speaker's name, gender, electoral district, role and party affiliation and the date of the speech, plus identifiers for protocols, speeches and speakers that link to further metadata in the SWERIK project and the Riksdag's open data. It is one of the largest open Swedish speech corpora and supports Swedish automatic speech recognition, speaker and political-speech analysis, and computational social science on parliamentary debate. This repository holds the dataset as Parquet shards under data/ (rixvox_0000.parquet onward, at least 944 files listed) with audio, text and metadata columns (measured 2026-11-14; the Hub reports about 2.7 TB). Stream with datasets.load_dataset('KBLab/rixvox-v2', streaming=True). Released under the Open Data Commons Attribution License (ODC-By 1.0); attribute RixVox-v2 and the Riksdag's open data.","displayTitle":"RixVox-v2: Swedish parliamentary speech dataset","id":"a5e8dfc9-5fa0-557d-bda5-30d9292ffc33","license":"ODC-By-1.0","licenseEvidenceUrl":"https://huggingface.co/datasets/KBLab/rixvox-v2","objectCountEstimate":946,"probeStatus":"ok","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"KBLab/rixvox-v2\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"KBLab/rixvox-v2\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=KBLab, license:odc-by, size_categories:1M<n<10M, format:parquet, downloads=5280"],"format":"mixed","generatedAt":"2026-11-14T14:15:23.949Z","layout":{"example":"data/rixvox_0000.parquet","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"data/rixvox_0000.parquet","truncatedAt":512}],"sampleObjects":["data/rixvox_0000.parquet","data/rixvox_0001.parquet","data/rixvox_0002.parquet","data/rixvox_0003.parquet","data/rixvox_0004.parquet"],"sidecars":[{"bytes":"---\nlicense: odc-by\ntask_categories:\n- automatic-speech-recognition\nlanguage:\n- sv\npretty_name: RixVox-v2\n---\n## RixVox-v2: A Swedish parliamentary speech dataset\n\nRixVox-v2 is a parliamentary speech dataset spanning nearly 23000 hours of speech. The dataset was built by matching and force aligning speeches in parliamentary protocols to media recordings of debates. Each observation contains metadata about the speaker's name, gender, district, role, party affiliation, and the date the speech was given. We include identifiers for protocols, speeches and speakers that allow linking observations in our dataset to further metadata in [SWERIK](https://swerik-project.github.io/) and [the Riksdag's open data](https://www.riksdagen.se/sv/dokument-och-lagar/riksdagens-oppna-data/). \n\n\n| Variable           | Description | \n| ------------------ | ----------- |\n| text               | Text from parliamentary protocols which have been [force aligned](https://pytorch.org/audio/main/tutorials/ctc_forced_alignment_api_tutorial.html) with audio. Protocols from 1966-2002 sourced from [SWERIK](https://github.com/swerik-project/the-swedish-parliament-corpus). Speeches from 2000-2024 sourced from [the Riksdag's open data](https://www.riksdagen.se/sv/dokument-och-lagar/riksdagens-oppna-data/anforanden/). |\n| audio              | 16 kHz, mono audio segments, up to 30 seconds in duration. |\n| name               | Speaker name. |\n| party              | Party membership(s) of the speaker over their career. |\n| gender             | Gender. |\n| role               | Role(s) or position(s) the speaker has had over their career. |\n| district           | The district(s) the person has represented over their career (available only if person has been a member of parliament). |\n| year               | The year the speech was given according to parliamentary protocol metadata. |\n| dates              | Date(s) when speech was possibly given according to parsed protocol metadata. A source protocol document can contain several debates spanning multiples dates. |\n| date_approx        | Earliest possible date speech could have been given according to `dates`. Included to simplify date based filtering. |\n| start              | Start time of this audio segment in the debate's audio file (seconds). See `audio_file` for source media filename. |\n| end                | End time of this audio segment in debate audio file (seconds) |\n| duration           | Duration of audio segment in seconds. |\n| audio_file         | Filename of the source media file in the Riksdag's media archive. |\n| text_normalized    | Normalized version of original protocol `text` used for calculating eval metrics (lowercased, alphanumeric, converted numbers to spelled out letters, no punctuation). |\n| text_timestamps    | Text with timestamps inserted at sentence boundaries. Enables training Whisper with ability to output timestamps. |\n| text_previous      | Text of the previous audio segment. Enables training Whisper with previous context as prompt. |\n| wav2vec_transcription | Audio transcribed by [wav2vec2](https://huggingface.co/KBLab/wav2vec2-large-voxrex-swedish). |\n| whisper_transcription | Audio transcribed by [Whisper-large-v3](https://huggingface.co/openai/whisper-large-v3). |\n| bleu_whisper       | BLEU score between `text_normalized` and normalized `whisper_transcription`. |\n| bleu_wav2vec       | BLEU score between `text_normalized` and normalized `wav2vec_transcription`. |\n| wer_whisper        | WER between `text_normalized` and normalized `whisper_transcription`. |\n| wer_wav2vec        | WER between `text_normalized` and normalized `wav2vec_transcription`. |\n| cer_whisper_first  | CER of 10 first characters of `text_normalized` and normalized `whisper_transcription`. |\n| cer_wav2vec_first  | CER of 10 first characters of `text_normalized` and normalized `wav2vec_transcription`. |\n| cer_whisper_last   | CER of 10 last characters of `text_normalized` and normalized `whisper_transcription`. |\n| cer_wav2vec_last   | CER of 10 last characters of `text_normalized` and normalized `wav2vec_tr","encoding":"utf8","key":"README.md","name":"README.md","size":5843,"truncated":true}],"sourceObjectKey":"data/rixvox_0000.parquet","starterCell":"from datasets import load_dataset\nds = load_dataset(\"KBLab/rixvox-v2\")\nprint(ds)","starterSamplePath":"data/rixvox_0000.parquet","totalObjects":946},"summary":"RixVox-v2 is a Swedish parliamentary speech dataset spanning nearly 23,000 hours of speech, built by KBLab at the National Library of Sweden by matching and force-aligning…","title":"RixVox-v2: Swedish parliamentary speech dataset","updatedAt":"2026-11-14T17:34:36.306Z","visibility":"public","websiteUrl":"https://huggingface.co/datasets/KBLab/rixvox-v2"}}