Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
audio
audioduration (s)
4.8
26.8
label
class label
102 classes
0af_za
1am_et
2ar_eg
3as_in
4ast_es
5az_az
6be_by
7bg_bg
8bn_in
9bs_ba
10ca_es
11ceb_ph
12ckb_iq
13cmn_hans_cn
14cs_cz
15cy_gb
16da_dk
17de_de
18el_gr
19en_us
19en_us
19en_us
20es_419
21et_ee
22fa_ir
23ff_sn
24fi_fi
25fil_ph
26fr_fr
26fr_fr
26fr_fr
27ga_ie
28gl_es
29gu_in
30ha_ng
31he_il
32hi_in
33hr_hr
34hu_hu
35hy_am
36id_id
36id_id
36id_id
37ig_ng
38is_is
39it_it
40ja_jp
40ja_jp
40ja_jp
41jv_id
41jv_id
41jv_id
42ka_ge
43kam_ke
44kea_cv
45kk_kz
46km_kh
47kn_in
48ko_kr
49ky_kg
50lb_lu
51lg_ug
52ln_cd
53lo_la
54lt_lt
55luo_ke
56lv_lv
57mi_nz
58mk_mk
59ml_in
60mn_mn
61mr_in
62ms_my
63mt_mt
64my_mm
65nb_no
66ne_np
67nl_nl
68nso_za
69ny_mw
70oc_fr
71om_et
72or_in
73pa_in
74pl_pl
75ps_af
76pt_br
77ro_ro
78ru_ru
79sd_in
80sk_sk
81sl_si
82sn_zw
83so_so
84sr_rs
85sv_se
86sw_ke
86sw_ke
86sw_ke
87ta_in
End of preview. Expand in Data Studio

FLEURS multilingual speech pilot (102 locale configurations, 116 clips)

This pilot is publicly hosted at https://huggingface.co/datasets/Kicaulah/fleurs-pilot-102. It was assembled as a sample for assessing FLEURS as one source in a broader multilingual audio dataset; its scope is intentionally broad across FLEURS language configurations and it is not statistically representative.

What's included

  • audio/: 116 WAV recordings across 102 locale configurations.
  • metadata/samples.csv: per-recording source row, locale, transcript fields, available gender label, technical audio properties, SHA-256 checksum, and direct manifest/archive provenance.
  • metadata/source_provenance.json: pinned source revision, version, language coverage, selection method, source links, and rights-review status.
  • sources/FLEURS_dataset_card_snapshot.md: dataset card saved from the pinned revision.
  • sources/FLEURS_citation.bib: recommended paper citation.
  • sources/CC-BY-4.0_legalcode.txt: official license text.
  • sources/rights_review.md: source-based notes on consent, privacy, publicity/personality rights, and reuse boundaries.
  • checksums.sha256: integrity hashes for package files (excluding this checksum file).

Language coverage and sampling

All 102 of 102 source language configurations are represented. Locale codes: af_za, am_et, ar_eg, as_in, ast_es, az_az, be_by, bg_bg, bn_in, bs_ba, ca_es, ceb_ph, ckb_iq, cmn_hans_cn, cs_cz, cy_gb, da_dk, de_de, el_gr, en_us, es_419, et_ee, fa_ir, ff_sn, fi_fi, fil_ph, fr_fr, ga_ie, gl_es, gu_in, ha_ng, he_il, hi_in, hr_hr, hu_hu, hy_am, id_id, ig_ng, is_is, it_it, ja_jp, jv_id, ka_ge, kam_ke, kea_cv, kk_kz, km_kh, kn_in, ko_kr, ky_kg, lb_lu, lg_ug, ln_cd, lo_la, lt_lt, luo_ke, lv_lv, mi_nz, mk_mk, ml_in, mn_mn, mr_in, ms_my, mt_mt, my_mm, nb_no, ne_np, nl_nl, nso_za, ny_mw, oc_fr, om_et, or_in, pa_in, pl_pl, ps_af, pt_br, ro_ro, ru_ru, sd_in, sk_sk, sl_si, sn_zw, so_so, sr_rs, sv_se, sw_ke, ta_in, te_in, tg_tj, th_th, tr_tr, uk_ua, umb_ao, ur_pk, uz_uz, vi_vn, wo_sn, xh_za, yo_ng, yue_hant_hk, zu_za.

The original seven locales retain three clips each; every other represented locale contributes one clip. Four extra clips came from the pinned validation-row API. For the remaining expansion, the first WAV member of each locale's pinned audio/dev.tar.gz archive was streamed and matched by filename to the pinned dev.tsv; this avoids downloading each full-language archive. See the per-file manifest/archive URLs and chosen member in metadata/samples.csv. In FLEURS, the source dev split corresponds to the validation split. This convenience sample is not statistically representative of speakers, accents, or dialects.

Total audio duration: 1444.4 seconds. All included clips are WAV; sample rate, channel count, bit depth, codec tag, frame count, duration, and SHA-256 are recorded in metadata/samples.csv. Audio files were copied without re-encoding.

Attribution and license

Recommended attribution:

FLEURS: Few-shot Learning Evaluation of Universal Representations of Speech, by Alexis Conneau, Min Ma, Simran Khanuja, Yu Zhang, Vera Axelrod, Siddharth Dalmia, Jason Riesa, Clara Rivera, and Ankur Bapna (2022), https://arxiv.org/abs/2205.12446. Source data: Google FLEURS, https://huggingface.co/datasets/google/fleurs, snapshot 70bb2e84b976b7e960aa89f1c648e09c59f894dd. The upstream dataset card labels FLEURS as Creative Commons Attribution 4.0 International (CC BY 4.0), https://creativecommons.org/licenses/by/4.0/.

This pilot consists of selected, unmodified source recordings organized by locale, with source metadata structured into a CSV. The package does not imply endorsement by Google, the paper authors, or Hugging Face.

Rights caution: The upstream Google FLEURS dataset card labels FLEURS CC BY 4.0, but the reviewed card and paper do not document speaker-consent terms or a per-speaker rights schedule. CC BY 4.0 does not grant privacy, publicity, or other personality rights. Google XTREME contest terms section 8.1 contain a use restriction, and their applicability to the separate FLEURS release is unresolved. FLEURS reuses FLoRes text, whose FLORES-101 card identifies CC BY-SA 4.0. This pilot is publicly hosted here at the owner's direction before written clarification; public availability does not establish speaker consent, rights, or permission for downstream redistribution or model training. Obtain written clarification about recording consent, audio/transcript license scope, rights, and XTREME terms before those downstream uses. See sources/rights_review.md.

Version and provenance

No dataset/audio was uploaded externally during assembly. The pilot was later uploaded to this public repo at the owner's direction on 2026-11-19, before written clarification.

Downloads last month
23

Paper for Kicaulah/fleurs-pilot-102