Subscribe
Datasets
verified 2026-11. Download on Hugging Face
126 of 126Human speech and open licences first
SaveFDLengthDomainLanguagesLicenceSourceSpeech
—42.6k h29.3k validated————295PermissiveConsentedhuman2019-06
—~3,500 hconv.————hi bn ta +19PermissiveConsentedhuman2024-03
CoVoST-2Meta AI
—2,880 h————21 + enPermissiveConsentedhuman2020-07
FLEURSGoogle Research
—~1k h————102+PermissiveConsentedhuman2022-05
Full-duplex280 h——Casual talk—enPermissiveConsentedhuman2026-02
AISHELL-1Beijing Shell Shell Technology
—178 h————zhPermissiveConsentedhuman2017-09
Full-duplex141 h——Casual talk—enPermissiveConsentedhuman2026-02
AMI Meeting CorpusUniversity of Edinburgh / Idiap
—100 h————enPermissiveConsentedhuman2006-06
AISHELL-3Beijing Shell Shell Technology
—85 h————zhPermissiveConsentedhuman2020-10
ICSI Meeting CorpusICSI / University of Edinburgh (AMI site)
—72 h————enPermissiveConsentedhuman2004-02
VCTK CorpusCSTR, University of Edinburgh
—44 h————enPermissiveConsentedhuman2019-11
Gridspace-Stanford Harper ValleyGridspace / Stanford CS224S
Full-duplex23.7 h59—Customer service—enPermissiveConsentedhuman2020-10
Full-duplex20 h—avg ~21 minTask-oriented—enPermissiveConsentedhuman2026-08
HCRC Map Task CorpusUniversity of Edinburgh / University of Glasgow (HCRC)
Full-duplex18 h64—Task-oriented—enPermissiveConsentedhuman—
Full-duplex17.4 h—avg ~22 minTask-oriented—enPermissiveConsentedhuman2026-09
PriMock57Babylon Health
—~9 h————enPermissiveConsentedhuman2022-05
—5.3 h————enPermissiveConsentedhuman2019-09
Full-duplex5.1 h8avg 6.8 minCustomer service—enPermissiveConsentedhuman2026-10
—~108 h————hi bn ta +19PermissiveMixedhuman2026-07
YODAS v3ESPnet / WavLab (CMU)
—1.1M h————100+PermissivePublic mediahuman2026-09
YODAS / YODAS2WavLab (CMU)
—500k h————100+PermissivePublic mediahuman2024-06
VoxPopuliMeta AI
—400k h————en de fr +20PermissivePublic mediahuman2021-01
Libri-LightMeta AI (FAIR)
—60k h————enPermissivePublic mediahuman2019-12
—50k h————en de nl +5PermissivePublic mediahuman2020-12
VoxLingua107Tallinn University of Technology
—6,628 h————multiPermissivePublic mediahuman2020-11
LibriSpeechVassil Panayotov et al.
—960 h————enPermissivePublic mediahuman2015-04
LibriTTSGoogle / Heiga Zen et al.
—585 h————enPermissivePublic mediahuman2019-04
—585 h————enPermissivePublic mediahuman2023-05
Hi-Fi TTSNVIDIA
—292 h————enPermissivePublic mediahuman2021-04
CSS10Kyubyong Park et al.
—140 h————de el es +7PermissivePublic mediahuman2019-03
MUSANJohns Hopkins University
—109 h————multiPermissivePublic mediahuman2015-10
LJSpeechKeith Ito
—24 h————enPermissivePublic mediahuman2017-07
Real-TurnTurkAta Technology Platforms
Full-duplex1.94 hpublic subset6avg ~19 min——trPermissiveNot statedhuman2026-08
Full-duplex128.6 h1565–15 minCustomer service—enShare-alikeConsentedhuman2026-04
AliMeetingAlibaba Group / OpenSLR SLR-119
—118.75 h————zhShare-alikeConsentedhuman—
CREMA-DUniversity of Pennsylvania
—~5 h————enShare-alikeConsentedhuman2014-08
—30k+ h————enShare-alikePublic mediahuman2021-11
—119 h————enShare-alikePublic mediahuman2022-03
—20 h————enShare-alikeConsentedhuman2022-07
GranaryNVIDIA
—643k h192k h with audio————en de fr +22PermissivePublic mediahuman2025-05
ReazonSpeechReazon Holdings
—35k h————jaShare-alikePublic mediahuman2023-03
VoxCeleb2VGG, University of Oxford
—2,442 h————multiShare-alikePublic mediahuman2018-06
VoxCeleb1VGG, University of Oxford
—352 h————enShare-alikePublic mediahuman2017-06
Full-duplex4,000+ h4,2842–10 minCasual talk, Task-oriented, Other—enNon-commercialConsentedhuman2025-06
SmoothConv / DuplexConvASLP@NPU / QualiaLabs
Full-duplex2,100 h100 h + 2,000 h—clips avg 1.3–2.4 minCasual talk, Other—zhNon-commercialConsentedhuman2026-06
—393 h————enNon-commercialConsentedhuman2021-06
MultiDialogIVLLab (Yonsei / Naver)
—340 h————enNon-commercialConsentedhuman2024-06
SpokenWOZAlibaba DAMO Academy
—249 h————enNon-commercialConsentedhuman2023-05
MagicData-RAMCMagic Data Technology / OpenSLR SLR-123
—180 h————zhNon-commercialConsentedhuman2022-03
AISHELL-5Beijing AISHELL Technology / OpenSLR SLR-159
—105 h————zhNon-commercialConsentedhuman2025-05
CASPERJohns Hopkins University (The SSSD Project)
Full-duplex102 h157avg 15.8 minCasual talk—enNon-commercialConsentedhuman2025-06
WHAM! / WHAMR!Whisper.ai / MERL
—~80 h————enNon-commercialConsentedhuman2019-07
CHiME-6CHiME Challenge
—40 h————enNon-commercialConsentedhuman2020-04
ExpressoMeta AI
—40 h————enNon-commercialConsentedhuman2023-08
Full-duplex15 h10 zh + 5 en14avg ~26 minCasual talk—zh enNon-commercialConsentedhuman2024-11
Full-duplex10.35 h————jaNon-commercialConsentedhuman2025-04
RAVDESSRyerson University
—~3 h————enNon-commercialConsentedhuman2018-04
Emilia / Emilia-LargeAmphion (CUHK-Shenzhen)
—215.6k hEmilia-Large————en zh ja +3Non-commercialPublic mediahuman2024-07
J-CHATSarulab (UTokyo) / Keio
—76k h————jaNon-commercialPublic mediahuman2024-07
WenetSpeechWeNet community / NPU-ASLP
—10k hlabelled————zhNon-commercialPublic mediahuman2021-10
MuST-CFBK (Fondazione Bruno Kessler)
—~4k h————en → 14Non-commercialPublic mediahuman2019-06
Multilingual TEDxJohns Hopkins / CMU
—760 h————es fr pt +5Non-commercialPublic mediahuman2021-02
TED-LIUM 3LIUM, Université du Mans
—452 h————enNon-commercialPublic mediahuman2018-05
CN-CelebCSLT, Tsinghua University
—274 h————zhNon-commercialPublic mediahuman2019-11
OleSpeech-IVOlewave / Stanford / Microsoft / PingAn / KU / CMU
—100 hopen subset————enNon-commercialPublic mediahuman2025-09
Full-duplex1,960 h11,917up to 10 minCasual talk$42,000$21/henGatedConsentedhuman2004-12
MLC-SLM (Interspeech 2025)Nexdata / Interspeech 2025 challenge
—1,604 h————en fr de +8GatedConsentedhuman2025-09
Full-duplex1,502 h2,645median 33 minCasual talk—en es fr +19GatedConsentedhuman2026-09
Open Yap 1KThe Agentic Data Company
Full-duplex1,000 h8.9 h open sample239avg 37.5 minCasual talk—enGatedConsentedhuman2026-09
AISHELL-2Beijing Shell Shell Technology
—1,000 h————zhGatedConsentedhuman2018-08
KsponSpeechETRI / Korean AI Hub
—969 h————koGatedConsentedhuman2020-03
SSSD (Scalable Spontaneous Speech Dataset)Carnegie Mellon University (WAVLab)
Full-duplex727 h209avg 25–30 minCasual talk—enGatedConsentedhuman2025-06
Corpus of Spontaneous Japanese (CSJ)NINJAL / NICT / Tokyo Tech
—661 h————jaGatedConsentedhuman2004-01
Full-duplex590 h——Casual talk$37,750$64/henGatedConsentedhuman2013-06
Full-duplex~313 h10 sets with stated hours—5–30 minCasual talk$39,750$127/hen fr ar +10GatedConsentedhuman—
Full-duplex260 h543avg ~6.5 minCasual talk$35,500$137/henGatedConsentedhuman1993-01
Full-duplex~229 h2026 Second Editions—up to 30 minCasual talk—ja zh es +3GatedConsentedhuman1996-01
Full-duplex149 h1,793up to 10 minCasual talk$36,000$242/hzhGatedConsentedhuman2005-07
Full-duplex104.9 h—avg 15 minCasual talk, Task-oriented, Other—enGatedConsentedhuman2026-06
—~80 h————enGatedConsentedhuman1993-01
DAI-ASR-I18NDavid AI
Full-duplex42.1 h682clips avg 2.4 minCasual talk—en zh ja +18GatedConsentedhuman2026-10
IEMOCAPUSC SAIL
—12 h————enGatedConsentedhuman2008-12
TIMITDARPA / LDC
—5.4 h———$34,125$6,319/henGatedConsentedhuman1993-10
GigaSpeechSpeechColab
—10k h————enGatedPublic mediahuman2021-06
SPGISpeechKensho (S&P Global)
—5,000 h————enGatedPublic mediahuman2021-04
MSP-PodcastUT Dallas (MSP Lab)
—409 hv2.0————enGatedPublic mediahuman2017-01
Full-duplex1,500 h150——$135,000$90/henCustomConsentedhuman—
Full-duplex1,125 h565median 15.5 minCasual talk$11,250$10/henCustomConsentedhuman2026-11
CANDORBetterUp Labs (Reece, Cooney et al.)
Full-duplex~860 h1,460median 29 minCasual talk—enCustomConsentedhuman2022-02
JVS CorpusTakamichi Lab (University of Tokyo)
—30 h————jaCustomConsentedhuman2019-08
—29 h————en zhCustomConsentedhuman2021-05
L2-ARCTICTexas A&M / Iowa State
—27 h————enCustomConsentedhuman2018-09
Dialogs (Russian)Shigabeev, Latyshev
—20.6 h————ruCustomConsentedhuman2026-07
JSUTShinnosuke Takamichi (U. Tokyo)
—10 h————jaCustomConsentedhuman2017-11
Full-duplex———Casual talk—enCustomConsentedhuman—
Full-duplex———Casual talk—enCustomConsentedhuman—
Full-duplex———Casual talk—enCustomConsentedhuman—
Full-duplex———Casual talk—koCustomConsentedhuman—
Full-duplex———Casual talk—jaCustomConsentedhuman—
Full-duplex—————jaCustomConsentedhuman—
Full-duplex—————jaCustomConsentedhuman—
Full-duplex———Task-oriented—enCustomConsentedhuman—
DuplexChatNakata, Saito, Saruwatari (U. Tokyo / AIST)
Full-duplex415k h—clips avg 1.1 minPodcasts and media—en jaCustomPublic mediahuman2026-07
MM-F2FXiamen University
—210 h————enCustomPublic mediahuman2025-07
MELDDeCLaRe Lab / SUTD
—~12 h————enCustomPublic mediahuman2018-10
Full-duplex305 h—10–60 minHealthcare, Meetings, Customer service$19,825$65/hzh en es +8CustomNot statedhuman2026-04
CVSSGoogle Research
—3,809 h————21 → enPermissiveConsentedmixed2022-01
NaijaS2STMaltais et al.
—~200 h————ig ha yo +2PermissiveConsentedmixed2026-04
——————ar bn zh +20PermissiveNot statedmixed2026-01
CVSS-XGris et al.
—~16,070 hsynthetic target speech————en +28Non-commercialConsentedmixed2026-09
SD-EvalAmphion (CUHK-Shenzhen)
—8.76 h————enNon-commercialMixedmixed2024-06
InteractSpeechInteractSpeech authors (EMNLP 2025)
—150 haudio not released————enCustomConsentedmixed2025-11
Semantic-VAD EOTScicom (MSC) Bhd
—4,064 hconversational configs————en fr de +12CustomMixedmixed2026-07
Easy Turn TrainsetASLP Lab, Northwestern Polytechnical University / Huawei
—1,145 h————en zhCustomMixedmixed2025-09
DuplexGenDuplexGen authors (Kim et al.)
—1,015 h————enPermissivePublic mediasynthetic2026-07
VoxDialogueVoxDialogue authors (ICLR 2025)
——————enPermissivePublic mediasynthetic2025-05
Behavior-SDSeoul National University (NAACL 2025)
—2,164 h————enPermissiveNot statedsynthetic2025-04
—preview only2,000.5 h claimed————koPermissiveNot statedsynthetic2026-08
VoiceAssistant-400Kgpt-omni (Mini-Omni authors)
——————enPermissiveNot statedsynthetic2024-09
VartalaapKapture CX
Full-duplex5,446 h—avg 1.8 minCustomer service, Casual talk—hi enNon-commercialConsentedsynthetic2026-08
InstructS2S-200KICTNLP (LLaMA-Omni team)
——————enNon-commercialPublic mediasynthetic2024-09
Full-duplex57.6k h————en zhNon-commercialNot statedsynthetic2026-09
—1,056 hEnglish side————en zhNon-commercialNot statedsynthetic2026-04
UltraVoiceSJTU X-LANCE / BIGAI (Tu et al.)
—833 h————en zh ja +1CustomConsentedsynthetic2025-10
FlexiSLM-DataAmphion team
—26.7k h————enCustomPublic mediasynthetic2026-08
DEMANDInria / University of Edinburgh
—~5 h————multiShare-alike——2013-06