SHACHI - Language Resource Metadata Database

The RWCP News Speech Database contains professionally read broadcast news data. The news scripts for reading were also written by a professional broadcast reporter based on the actual event. Each announcer read a total of 40 stories, totaling about an hour, and then Set A (50 sentences) of ATR phonetically balanced sentences.

The RWCP Meeting Speech Corpus contains speech data of simulated meetings by three or more participants. The subjects of meeting includes; package tour plan, making a website of a travel agent, mail magazine planning of a travel agent, and research on people's opinion about travel.

The database contains; "RWCP Sound Scene Database in Real Acoustical Environments" (Vol.1), speech data of fixed sound sources measured be microphone-array (Vol.2), and speech data of moving sound sources measured by microphone-array (Vol.3).

CIAIR-VCV is a collection of 288 elementary school students' voices (words and sentences) inside a room in a normal living environment.

CENSREC-1(AURORA-2J) is a noisy digit speech database and a Japanized version of the AURORA-2 database. "CENSREC" stands for "Corpus and Environments for Noisy Speech RECognition. The CENSREC-1 database was created in exactly the same way as the AURORA-2 database, but it was uttered in Japanese, like "ichi, ni, san," for "one, two, three" in AURORA-2. The number of speakers is the same, and the digit strings for each speaker are identical as the AURORA-2, too.

CENSREC-1-C is a speech database for the evaluation of voice activity detection in several noise environment. The simulated speech data of CENSREC-1-C are constructed by concatenating several utterances spoken by one speaker. The vocabulary of simulated data included in the CENSREC-1-C consist of eleven Japanese digits ("ichi," "ni," "san," ...), which are the same as in CENSREC-1(AURORA-2J). The noise environments for simulated data are the same as CENSREC-1, too.

CENSREC-2 is a database for the evaluation of continuous digit recognition in real car driving environments. The digit sequence of each utterance and the pronunciation of Japanese digits in this database are the same as the CENSREC-1(AURORA-2J) database. The data were recorded under 11 environmental conditions using combinations of three kinds of vehicle speeds and four kinds of in-car environments.

CENSREC-3 is an in-car speech database for the evaluation of isolated word recognition in real driving car environments. The data was recorded under 16 environmental conditions using combinations of three kinds of vehicle speeds and six kinds of in-car environments . For training data, driver's speech of phonetically-balanced sentences was recorded. 50 words were recorded as testing data basically for each person in each environment.

The UME-ERJ corpus was made in support of the Priority Areas Project on "Advanced Utilization of Multimedia to Promote Higher Education Reform" during 2000-2002. It is an English speech database read by Japanese undergraduate and graduate students. Grading lists by native English teachers are also included in the corpus.

The UME-JRF corpus was made in support of the Priority Areas Project on "Advanced Utilization of Multimedia to Promote Higher Education Reform" during 2000-2002. It is a Japanese speech database read by foreign undergraduate and graduate students studying in Japan. Grading lists by native Japanese teachers are also included in the corpus.