Datasets

This page aims to provide a list of datasets with detailed attributes and links to corresponding research results (papers, numerical results, output transcriptions, intermediary data, etc). Each dataset may be used for one or more applications: automatic speech recognition, speaker identification and verification, source localization, speech enhancement and separation...

Disclaimer: Only publicly available datasets with a total duration longer than 5 min are listed.

Datasets	General attributes								Speech							Channel				Noise	Ground truth
	release	scenario	total duration	sampling rate	degraded channels	cameras	cost	links	speech duration	unique speakers	language	unique words	speaking style	simultaneous speakers	speaker overlap	channel type	radiation	speaker location	speaker movements	noise type	speech signal	speaker location and orientation	words	nonverbal traits	noise events
ShATR	1994	meeting	37 min	48000	3 (distant)	no	free	download email paper	37 min	5	UK English	1k	colloquial	5	multiple conversations	reverb	human	quasi-fixed	head	meeting	headset	yes	yes	no	yes
LLSEC	1996	conversation	1.4 h	16000	4 (distant)	no	free	download email	?	12	N/S	N/S	read, colloquial	2	conversation	reverb	human	quasi-fixed	head	hallway, restaurant	no	yes	no	no	no
RWCP Spoken Dialog Corpus	1996 - 1997	conversation	10 h	16000	2 (close but cross-talk)	no	free	download email paper	10 h	39	Japanese	?	colloquial	1 or 2	conversation	reverb	human	quasi-fixed	head	stationary background noise	no	no	yes	no	no
Aurora-2	2000	public spaces	33 h	8000-16000	1 (close)	no	TIDigits	download email paper	33 h	214	US English	11	digits	1	no	no (simulated telephone channel)	human	N/S	no	various real environments	original	N/S	yes	no	yes
SPINE1, SPINE2	2000 - 2001	military	38 h	16000	2 (close)	no	2 x ($800 (audio) + $500 (transcripts)) + 3 x ($1000 (audio) + $600 (transcripts))	purchase email paper	?	100	US English	1k	command, colloquial	1 or 2	no	no (simulated transmission channels)	human	quasi-fixed	head	military (pre-recorded noise played in sound booth while recording speech)	no	no	yes	no	no
Aurora-3 (subset of SpeechDat-Car)	2000 - 2003	car	?	16000	3 (+1 GSM) (distant)	no	1000 €	purchase papers	?	?	Finnish, German, Spanish, Danish, Italian	?	digits, command, read, spontaneous	1	no	reverb	human	quasi-fixed	head	car	close-talk	no	yes	no	no
RWCP Meeting Speech Corpus	2001	meeting	3.5 h	16000 - 48000	1 (distant)	3	free	download email paper	3.5 h	?	Japanese	?	colloquial	1 to 5	meeting	low reverb	human	quasi-fixed	head	stationary background noise	headset	no	yes	no	no
RWCP Real Environment Speech and Acoustic Database	2001	domestic, office	?	16000 - 48000	30 (distant)	no	free	download email paper	?	5	Japanese	?	read	1	no	real rir, reverb	loudspeaker	various	no, pivoting arm	stationary background noise	original	yes	yes	no	yes
SpeechDat-Car	2001 - 2011	car	?	16000	3 (+1 GSM) (distant)	no	1.1 Million for all 10 languages. Each costs 39k to 182k	purchase paper	?	300 per lang	Multiple	?	digits, command, read, spontaneous	1	no	reverb	human	quasi-fixed	head	car	close-talk	no	yes	no	no
Aurora-4	2002	public spaces	?	8000 - 16000	1 (close)	no	WSJ0	download email paper	?	101	US English	10k	read	1	no	no (simulated telephone channel)	human	N/S	no	various real environments	original	N/S	yes	no	yes
TED	2002	seminar	47 h	16000	1 (distant)	no	$275 (audio) + $250 (transcripts)	purchase paper	47 h	188	English (mostly non-native)	?	lecture	1 or more	seminar	reverb	human	quasi-fixed	head	stationary background noise	lapel	no	yes (partial)	no	no
CUAVE	2002	cocktail party	3 h	44100	1 (distant)	1	free	download email paper	3 h	36	US English	10	digits	1 or 2	full	reverb	human	quasi-fixed	head	stationary background noise	no	no	yes	no	no
CU-Move ("Microphone Array Data"; downsampled data with more speakers but less channels exist)	2002 - 2011	car	286 h	44100	6 to 8 (distant)	no	$25k with UT-Drive	purchase email paper	286 h	172	US English	12k	digits, command, read, dialogue	1	no	reverb	human	quasi-fixed	head	car	no	no	yes	no	no
CENSREC-1 (Aurora-2J)	2003	public spaces	?	8000	1 (close)	no	free	download email paper		214	Japanese	11	digits	1	no	various microphones and simulated channels	human	N/S	no	various real environments	original	N/S	yes	no	yes
AVICAR	2004	car	29 h	16000	7 (distant)	4	free	download email paper	29 h	86	US English, non-native English	1k	read	1	no	reverb	human	quasi-fixed	head	car	no	no	yes	no	no
AV16.3	2004	meeting	1.5 h	16000	16 (distant)	3	free	download email paper	1.5 h	12	N/S	N/S	colloquial	1 to 3	full	reverb	human	various	walk	stationary background noise	no	yes	no	no	no
ICSI Meeting Corpus	2004	meeting	72 h	16000	6 (distant)	no	$1900 (audio) + $900 (transcripts)	purchase email paper	72 h	53	US English	13k	meeting	3 to 10	meeting	reverb	human	quasi-fixed	head	stationary background noise	headset (some lapel)	no	yes	yes	no
NIST Meeting Pilot Corpus Speech	2004	meeting	15 h	16000	7 (distant)	no (released but not currently available for download)	$4000 (audio) + $1500 (transcripts)	purchase email paper	15 h	61	US English	6k	meeting	3 to 9	meeting	reverb	human	various	walk	stationary background noise	headset+lapel	no	yes	no	no
CHIL Meetings	2004 - 2007	seminar, meeting	60 h	44100	79 to 147 (distant)	6 to 9	3 500	purchase email paper	?	?	non-native English	?	seminar, meeting	3 to 20	seminar, meeting	reverb	human	quasi-fixed	head	meeting (scenarized)	headset	yes	yes	yes	no
SPEECON	2004 - 2011	public space, domestic, office, car	?	16000	3 (distant)	no	29 x 75000 for all languages	purchase email paper	?	600 per lang	Multiple	?	command, read, spontaneous	1	no	reverb	human	quasi-fixed	head	various real environments	headset	no	yes	no	no
CENSREC-2	2005	car	?	16000	1 (distant)	no	free	download email paper	?	214	Japanese	11	digits	1	no	reverb	human	quasi-fixed	head	car	headset	no	yes	no	no
CENSREC-3	2005	car	?	16000	1 (distant)	no	free except phonetically balanced training set: JPY 21000 (Universities) / JPY 105000 (Companies)	purchase email paper	?	18 (+293 in training)	Japanese	50 in evaluation; unknown but larger in phonetically-balanced utterances of training set	read	1	no	reverb	human	quasi-fixed	head	car	headset	no	yes	no	no
Aurora-5	2006	public spaces, domestic, office, car	?	8000	1 (distant)	no	TIDigits	download email paper	?	225	US English	11	digits	1	no	no, simulated, real rir	loudspeaker	N/S	no	various real environments	original	no	yes	no	yes
AMI	2006	meeting	100 h	16000	16 (distant)	6	free	download email paper	?	189	UK English	8k	meeting	4 (18% overlap)	meeting	reverb	human	quasi-fixed	head	stationary background noise	headset+lapel	yes	yes	yes	no
PASCAL SSC	2006	cocktail party	18.5 min (+ 8.5h clean training data)	25000	1 (mixing console)	no	free	email paper	18.5 min (+ 8.5h clean training data)	34	UK English	51	command	2	full	no	human	N/S	no	no	original	N/S	yes	no	no
HIWIRE	2007	airplane	21 h	16000	1 (close)	no	50	purchase email paper	21 h	81	non-native English	133	command	1	no	no	human	N/S	head	airplane	original	N/S	yes	no	no
UT-Drive	2007	car	40 h	25000	5 (distant)	2	$25k with CU-Move	download email paper	40 h	25 (more exist but not included in latest release 3.0)	US English	2.4k (but transcription is incomplete)	command, dialogue	1 to 2	conversation	reverb	human	quasi-fixed	head	car	headset (low quality)	no	yes (partial)	no	no
SASSEC, SiSEC underdetermined	2007 - 2011	cocktail party	19 min	16000	2 (distant)	no	free	download email paper	19 min	16	N/S	N/S	read	3 or 4	full	simulated, real rir, reverb	no, loudspeaker	fixed	no	no	original+spatial image	yes	no	no	no
MC-WSJ-AV, PASCAL SSC2, 2012_MMA, REVERB RealData	2007 - 2014	cocktail party	10 h	16000	8 to 40 (distant)	no	$1 500	purchase email paper paper	?	45	UK English	10k	read	1 or 2	full	reverb	human	various	walk	stationary background noise	headset+lapel	yes	yes	no	no
CENSREC-4 (Simulated)	2008	public spaces, domestic, office, car	?	16000	1 (distant)	no	free	download email paper	?	214	Japanese	11	digits	1	no	real rir	mouth simulator	fixed	no	various real environments	original	no	yes	no	yes
CENSREC-4 (Real)	2008	public spaces, domestic, office, car	?	16000	1 (distant)	no	free	download email paper	?	10	Japanese	11	digits	1	no	reverb	human	quasi-fixed	head	various real environments	headset	no	yes	no	yes
DICIT	2008	domestic	6 h	48000	16 (distant)	2	free	download email paper	1 h	?	Italian	?	command	4	no	reverb	human	various	walk	domestic (scenarized)	headset+tv	yes	yes	no	yes
SiSEC head-geometry	2008	cocktail party	1.9 h	16000	2 (distant)	no	free	download email paper	1.9 h	?	N/S	N/S	read	2	full	real rir	loudspeaker	various	no	no	original+spatial image	yes	no	no	no
COSINE	2009	conversation	38 h	48000	20 (distant)	no	free	download email paper	11 h	91	US English, non-native English	5k	colloquial	2 to 7	conversation	reverb	human	various	walk	various real environments	headset+throat mic	no	yes	no	no
SiSEC real-world noise	2010	public spaces	20 min	16000	2 to 4 (distant)	no	free	download email paper	20 min	6	N/S	N/S	read	1 or 3	full	no, reverb (other room)	loudspeaker	various	no	various real environments	original+spatial image	yes	no	no	no
SiSEC dynamic	2010 - 2011	cocktail party	11 min	16000	2 to 4 (distant)	no	free	download email paper	11 min	?	N/S	N/S	read	Many but only 2 simultaneous	simu	reverb	loudspeaker	various	simu	no	original+spatial image	yes	no	no	no
CHiME 1, CHiME 2 Grid	2011 - 2012	domestic	70 h with some overlap	16000	2 (distant)	no	free	download email paper	12 h	34	UK English	51	command	1	no	real rir	dummy	quasi-fixed	simu	domestic	yes	yes	yes	no	no
CHiME 2 WSJ0	2012	domestic	78 h with some overlap	16000	2 (distant)	no	WSJ0	download email paper	33 h	101	US English	11k	read	1	no	real rir	dummy	fixed	no	domestic	yes	yes	yes	no	no
ETAPE	2012	debates, outdoor interviews, and other TV/radio broadcasts selected for large speaker overlap and/or noise	42 h	16000	1 (mixing console)	1	?	email paper	32 h	347	French	16k	colloquial	1 or more (7% overlap on average, up to 10% in debates)	conversation	some reverb	human	quasi-fixed	head	various real environments	no	N/S	yes	no	yes
GALE (Chinese broadcast conversation)	2013	conversation (TV Broadcast)	120 h	16000	1 (mixing console)	no	$2000 (audio) + $1500 (transcripts)	purchase email	108 h	?	Mandarin	?	colloquial	1 or more	conversation	no	human	quasi-fixed	head	no	no	N/S	yes	no	no
GALE (Arabic broadcast conversation)	2013	conversation (TV Broadcast)	251 h	16000	1 (mixing console)	no	2 x [$2000 (audio) + $1500 (transcripts)]	purchase email	234 h	?	Arabic	?	colloquial	1 or more	conversation	no	human	quasi-fixed	head	no	no	N/S	yes	no	no
REVERB SimData	2013	domestic, office	25 h	16000	8 (distant)	no	WSJCAM0	purchase email paper	25 h	130	UK English	10k	read	1	no	real rir	loudspeaker	fixed	no	experimental room	original+spatial image	yes	yes	no	yes
DIRHA	2014	domestic	3.8 h	48000	40 (distant)	no	free	download email paper	1.3 h	30	Italian, German, Greek, Portuguese	various	various	1 or more	simu	real rir	loudspeaker	various	no	domestic (sum of individual noises)	yes	yes	yes	no	yes

Automatic speech recognition

1st CHiME Challenge (2011)

Artificially distorted version of the small vocabulary GRID audio-visual corpus (audio only). Binaural reverberated speech with speaker situated in front of the microphones. Additive household noises impinging from different directions. Clean-training, noisy-training, development and evaluation sets available, see

Jon Barker, E. Vincent, N. Ma, H. Christensen, P. Green, "The PASCAL CHiME speech separation and recognition challenge", Computer Speech & Language, Volume 27, Issue 3, May 2013, Pages 621-633.

Available from Computer Speech and Language here

Corpus available here (no cost)

Resources

• Training recipe of the challenge for HTK here.

Baselines

• See the paper above for results for a wide range of techniques.

AURORA 5 (2007)

Artificially distorted version of the digits TI-DIGITS corpus. Additive noise and additive noise plus reverberant speech sets. Variable SNR range. Various mixed training sets, no evaluation set, see

G. Hirsch "Aurora-5 Experimental Framework for the Performance Evaluation of Speech Recognition in Case of a Hands-free Speech Input in Noisy Environments", Niederrhein University of Applied Sciences, 2007.

Paper available online here (no cost)

Corpus available from LDC here

Resources

• Training recipe for HTK is provided with the corpora.

Baselines

• Reproducible baseline: The above cited paper includes a baseline for the ETSI Advanced Front-End.

AURORA 4 (2002)

Artificially distorted version of the 5K word Wall Street Journal corpus (WSJ0). Stationary and non-stationary noises added. Second recordings with distant mismatched microphone. Clean-training, mixed-training, noisy training and test sets available. No evaluation set, see

G. Hirsch "Experimental Framework for the Performance Evaluation of Speech Recognition Front-ends on a Large Vocabulary Task", ETSI STQ Aurora DSR Working Group, 2002.

Paper available with the corpus.

Corpora available from ELRA here and here

Resources

• Training recipe for HTK available here. Note that this recipe is for Wall-Street Journal (WSJ0), which is the clean speech version of AURORA4. Small changes are needed in the feature extraction scripts to account for different file terminations.

Contribute a dataset

To contribute a new dataset, please

• create an account and login
• go to the wiki page above corresponding to your application; if it does not exist yet, you may create it
• click on the "Edit" link at the top of the page and add a new section for your dataset (the datasets are ordered by year of collection)
• click on the "Save page" link at the bottom of the page to save your modifications

Please make sure to provide the following information:

• name of the dataset and year of collection
• authors, institution, contact information
• link to the dataset and to side resources (lexicon, language model, etc)
• short description (nature of the data, license, etc) and link to a paper/report describing the dataset, if any
• at least 1 research result obtained for this dataset (see below)

We currently cannot provide storage space for large datasets. Please upload the dataset at a stable URL on the website of your institution or elsewhere and provide its URL only. If this is not possible, please contact the resources sharing working group.

Contribute a research result

To contribute a new research result, please

• create an account and login
• go to the wiki page and the section corresponding to the dataset for which this result was obtained
• click on the "Edit" link on the right of the section header and add a new item for your result
• click on the "Save page" link at the bottom of the page to save your modifications

Please make sure to provide the following information:

• authors, paper/report title, means of publication
• link to the pdf of the paper
• link to derived data (output transcriptions, intermediary data, etc)
• Code and instructions to reproduce experiments (if available)

In order to save storage space, please do not upload the paper on this wiki, but link it as much as possible from your institutional archive, from another public archive (e.g., arxiv) or from the publisher website (e.g., ieexplore).

We currently cannot provide storage space for large datasets. Please upload the derived data at a stable URL on the website of your institution or elsewhere and provide its URL only. If this is not possible, please contact the resources sharing working group.