SHACHI - Language Resource Metadata Database

The Aurora project was originally set up to establish a world wide standard for the feature extraction software which forms the core of the front-end of a DSR (Distributed Speech Recognition) system. ETSI formally adopted this activity as work items 007 and 008.The two work items within ETSI are :

- ETSI DES/STQ WI007 : Distributed Speech Recognition - Front-End Feature Extraction Algorithm & Compression Algorithm
- ETSI DES/STQ WI008 : Distributed Speech Recognition - Advanced Feature Extraction Algorithm.

This database is a subset of the SpeechDat-Car database in Spanish language which has been collected as part of the European Union funded SpeechDat-Car project. It contains isolated and connected Spanish digits spoken in the following noise and driving conditions inside a car :

1. Quiet environment. Stop motor running.
2. Low noise. Town traffic + low speed rough road.
3. High noise : High speed good road.

Written Corpora
The An-Nahar Lebanon Newspaper Text Corpus comprises articles in standard Arabic from 1995 to 2000 (6 years) stored as HTML files on CDRom media. Each year contains 45 000 articles and 24 million words. Each article includes information such as title, newspaper's name, date, country, type, page, etc. For each year, the size in byte is as follows:
1995 : 128 MB
1996 : 138 MB
1997 : 152 MB
1998 : 140 MB
1999 : 130 MB
2000 : 118 MB

Desktop/Microphone
The BABEL Database is a speech database that was produced by a research consortium funded by the European Union under the COPERNICUS programme (COPERNICUS Project 1304). The project began in March 1995 and was completed in December 1998. The objective was to create a database of languages of Central and Eastern Europe in parallel to the EUROM1 databases produced by the SAM Project (funded by the ESPRIT programme).
The BABEL consortium included six partners from Central and Eastern Europe (who had the major responsibility of planning and carrying out the recording and labelling) and six from Western Europe (whose role was mainly to advise and in some cases to act as host to BABEL researchers). The five databases collected within the project concern the Bulgarian, Estonian, Hungarian, Polish, and Romanian languages.
The Estonian database consists of the basic "common" set which is:
- Many Talker Set: 30 males, 30 females; each to read 50 numbers, 1-2 connected passages, 1 block of "filler" sentences, and 1 block of syllables.
- Few Talker Set: 4 males, 4 females; each to read 50 numbers, 10 connected passages, 1 block of "filler" sentences, and 2-3 blocks of syllables.
- Very Few Talker Set: 1 male, 1 female; each to read 2 blocks of 50 numbers, 40 connected passages, 4 blocks of "filler" sentences, and 9 blocks of syllables.
And the extension part: a short description of Estonian sound system.

Desktop/Microphone
The BABEL Romanian Database is a speech database that was produced by a research consortium funded by the European Union under the COPERNICUS programme (COPERNICUS Project 1304). The project began in March 1995 and was completed in December 1998. The objective was to create a database of languages of Central and Eastern Europe in parallel to the EUROM1 databases produced by the SAM Project (funded by the ESPRIT programme).

The BABEL consortium included six partners from Central and Eastern Europe (who had the major responsibility of planning and carrying out the recording and labelling) and six from Western Europe (whose role was mainly to advise and in some cases to act as host to BABEL researchers). The five databases collected within the project concern the Bulgarian, Estonian, Hungarian, Polish, and Romanian languages.

The Romanian database consists of the basic "common" set which is:

* The Many Talker Set: 50 males, 50 females; each to read 4 connected passages, 1 block of 2-3 "filler" sentences, 4 phonemically compact sentences, 3-7 individual sentences, and 26 numbers.
* The Few Talker Set: 5 males, 5 females from the Many Talker Set; each to read additionally 3 blocks of syllables and, in 4 supplemental sessions, 16 connected passages, 4 blocks of 2-3 "filler" sentences, 4 repetitions of the 26 numbers.
* The Very Few Talker Set: 1 male, 1 female from the Few Talker Set; each to read additionally 5 pairs of context words and the syllables in these 5 contexts.

Desktop/Microphone
A French speech database dedicated to the study of the perturbations of speech production due to noisy environments, and especially the Lombard effect. Environment: 4 noise conditions and the reference condition (quiet). The 2 noises used (a "white noise" and a "cocktail-party noise") were both produced to the ears of the speaker first without then with the feedback of his own voice. 10 speakers: 5 male and 5 female (4 CDROMs, approximately 2.2 Gigabytes)
Corpora:
· 30 Sentences (among the 50 "patching sentences" from EUROM1)
· 30 Numbers: digits (0 - 9) and 20 numbers ( 0 - 9999) from the fifth number group of EUROM1.
· "Logatomes" : 82 CVC with variation of initial and final consonant and the vowels [a,i,u] (EUROM1 set).

Desktop/Microphone
The BDSONS Database is a French - speech database with two subsets: evaluation and acoustic modelling. The Corpora consist of 32 speakers: 16 male and 16 female (7 CD-ROMs of approximately 3,5 Gigabytes), Phonetic labelling (partly) available on additional floppies, of the following data:
"Evaluation" (32 speakers): adjustment: 5 sentences and 54 bi-syllabic "logatomes", numbers, digits, letters, and names (spelled in isolation and in connected speech).
"Acoustic" (12 speakers): Words: 600 CVCV including 20 consonant and semi-consonant and vowels /a/, /i/, /u/ ; 200 consonant clusters; rhyme tests for consonant and vowels (pairs and triplets), sentences: 52 phonetically balanced sentences, 44 nasal sentences, 192 sentences including real words in French with 16 consonants and 12 vowels. Phonetic labelling for a subset of the data is available on floppy disk.

Desktop/Microphone
BREF-120 resulted from the efforts of LIMSI-CNRS researchers under sponsorship from the GDR-PRC CHM, the ACCT (OFIL), the EEC (ESPRIT Polyglot project), and the Aupelf-Uref.
A sub-set of BREF-120 is BREF-80 (ELRA-S0006), which consists of about 50-60 sentences per speaker and recordings conducted only with a Shure microphone. In BREF-80, the sentences were chosen to cover as many prompts as possible.
The BREF-120 corpus was designed to provide read speech data for the development and evaluation of continuous speech recognition systems (both speaker-dependent and speaker-independent), and to provide a large corpus of continuous speech for the acquisition of acoustic-phonetic knowledge of spoken French.
BREF-120 is a large read-speech corpus containing over 100 hours of speech material, from 120 speakers (55 males and 65 females). The text materials were selected verbatim from extracts of the French newspaper "Le Monde". Each of 80 speakers read approximately 10,000 words (about 650 sentences) of text, and another 40 speakers each read about half that amount. Simultaneous recordings were made in a sound-proof room using a Shure SM10 microphone and a Crown PCC160 microphone and were monitored to assure their contents. The speech signal was sampled at 16 kHz and digitised with 16 bits. The BREF-120 corpus contains 28 CDs; numbers 1-13 contain the Shure recorded data and numbers 14-28 contain the Crown recorded data

Desktop/Microphone
The BREF corpus was designed to provide enough read speech data for the development and evaluation of continuous speech recognition systems (both speaker-dependent and speaker-independent), and to provide a large corpus of continuous speech for the acquisition of acoustic-phonetic knowledge of spoken French. All the recorded texts were selected from extracts of the French newspaper Le Monde so as to provide a large vocabulary (over 20,000 words) and a wide range of phonetic environments. The entire BREF corpus contains over 100 hours of speech material from 120 speakers.
The BREF-80 sub-corpus consists of 2 ISO9660 CDROMs, BREF80-1 and BREF80-2, containing speaker-independent training data from 80 speakers. Together these 2 CDs contain 5330 sentences, an average of 67 sentences per speaker. While this data represents only a small portion of the entire BREF corpus, the sentences have been selected to cover most of the BREF training prompts, in order to conserve a wide range of phonetic contexts with a minimum amount of speech data. Thus, the BREF80 sub-corpus produced on these CDs was especially selected to train speaker-independent, vocabulary-independent speech recognizers.

Desktop/Microphone
The BREF-Polyglot is a sub-corpus of the BREF corpus (1 ISO9660 CDROM); it contains speaker-dependent training data from 6 speakers. There are a total of 3193 sentences (2 signal files for each sentence), on average 530 per speaker. While this data represents only a small portion of the entire BREF corpus, the sentences have selected to cover a wide range of phonetic contexts with a minimum amount of speech data. The BREF-Polyglot sub-corpus has been selected to train speaker-dependent, vocabulary-independent speech recognizers.

Telephone
The Basque FDB-1060 database contains the recordings of 1,060 speakers (480 males, 580 females) of Basque recorded over the fixed telephone network. This database is partitioned into 4 CDs. The database complies with the common specifications created in the SpeechDat project.

Speech samples are stored as sequences of 8-bit 8 kHz A-law. Each prompted utterance is stored in a separate file. Each signal file is accompanied by an ASCII SAM label file which contains the relevant descriptive information.

Each speaker uttered the following items:

* 6 common application words
* 1 sequence of isolated digits
* 4 digit strings : prompt sheet number, telephone number, credit card number, PIN code
* 3 dates : spontaneous, date (birth date), Prompted date (word style), relative and general date expr.
* 1 application word phrase
* 1 isolated digit
* 3 spelled word : spontaneous, spelled own forename, spelled directory city name, spelled artificial words
* 1 money amount
* 1 natural number
* 5 directory assistance: forename (spontaneous), city of origin (spontaneous), country name (most frequent city), most frequent company/agency name, forename & surname
* 2 spontaneous yes/no questions
* 9 phonetically rich sentences
* 2 time phrases : time of day (spontaneous), time phrase
* 4 phonetically rich words

The following age distribution has been obtained: 8 speakers are under 16, 474 are between 16 and 30, 320 are between 31 and 45, 236 are between 46 and 60, and 13 speakers are over 60. (The age of 9 speakers was not determined.)

A pronunciation lexicon with a phonemic transcription in SAMPA is also included.