SHACHI - Language Resource Metadata Database

Desktop/Microphone
This database contains the recordings of one male Catalan professional speaker recorded in a noise-reduced room simultaneously through a close talk microphone, a mid distance microphone and a laryngograph signal. This database consists in the recordings and annotations of read text material of approximately 10 hours of speech for baseline applications (Text-to-Speech systems). The FESTCAT Catalan TTS Baseline Male Speech Database was created within the scope of the FESTCAT project, funded by the Catalan Government.

Desktop/Microphone
This database contains the recordings of one female Catalan professional speaker recorded in a noise-reduced room simultaneously through a close talk microphone, a mid distance microphone and a laryngograph signal. It consists in the recordings and annotations of read text material of approximately 10 hours of speech for baseline applications (Text-to-Speech systems). The FESTCAT Catalan TTS Baseline Female Speech Database was created within the scope of the FESTCAT project funded by the Catalan Government.

Desktop/Microphone
This database contains the recordings of four female and four male Catalan professional speakers recorded in a noise-reduced room simultaneously through a close talk microphone, a mid distance microphone and a laryngograph signal. It consists of the recordings and annotations of read text material of approximately 1 hour of speech per speaker for baseline applications (Text-to-Speech systems). The FESTCAT Catalan TTS baseline speech database - 8 speakers was created within the scope of the FESTCAT project funded by the Catalan Government.

Desktop/Microphone
This database contains the Festival HTS models trained with 10h of speech from the TC-STAR Spanish Baseline Male Speech Database (ELRA-S0310).

Desktop/Microphone
This database contains the Festival HTS models trained with 10h of speech from the TC-STAR Spanish Baseline Female Speech Database (ELRA-S0309).

Desktop/Microphone
This database contains Bilingual (English and Spanish) Festival HTS models. Models were trained with 9h of speech from 2 female bilingual speakers and 2 male bilingual speakers. Each speaker recorded 2h 15 min per language. The speech data can be found in the TC-STAR Bilingual Voice-Conversion Spanish Speech Database (ELRA-S0311) and in the TC-STAR Bilingual Expressive Spanish Speech Database (ELRA-S0313).

Desktop/Microphone
This database contains a unit-selection voice (clunits technology) for their use in Festival Synthesis System (tested on version 2.0.95:beta April 2010).
The voice was built using a subset of speech derived from the TC-STAR Spanish Baseline Male Speech Database: mid distance microphone, 2h26m, 16kHz, 16bits. The complete speech data (10hours, 96kHz, 24 bits, 3 channels) can be found in the TC-STAR Male Baseline Voice Spanish Database, which is also available in the ELRA catalogue (ELRA-S0310) and Meta-Share. The Spanish Festival voice male was created within the scope of the METANET4U project funded by the European Commission.

Desktop/Microphone
This database contains a unit-selection voice (clunits technology) for their use in Festival Synthesis System (tested on version 2.0.95:beta April 2010).
The voice was built using a subset of speech derived from the TC-STAR Spanish Baseline Female Speech Database: mid distance microphone, 4h25m, 16kHz, 16bits. The complete speech data (10hours, 96kHz, 24 bits, 3 channels) can be found in the TC-STAR Female Baseline Voice Spanish Database, which is also available in the ELRA catalogue (ELRA-S0309) and Meta-Share. The Spanish Festival voice - female was created within the scope of the METANET4U project funded by the European Commission.

Telephone
The speech corpus VERIF1DE contains 20 recordings (sessions) of 150 German speakers each over the telephone network. Sessions are distributed over a period of time (speakers were not allowed to record more than one session a day, but this was not controlled). Each session contains 40 single recordings, mainly speech read from a prompt sheet. The acoustic environment is classified into quiet (14) and noisy (6); the telephone channel is either fixed network (10) or GSM (10). Recorded content is mainly SpeechDat conform extended by some extra items. The corpus has been pre-validated by BAS, then repaired (extra recordings to replace missing items), and then a final validation was performed by SPEX. After the final validation (version 1) the corpus was again repaired to some extent (version 1.3).

Content of version 1.3, dated 24/10/2001:
• Number of speakers: 150
• Number of sessions: 3,000
• Number of recordings: 120,000
• File formats:
- SpeechDat European telephone format ALAW 8kHz, 8 bit, raw
- SpeechDat annotation files (SAM)
- Metadata: speaker and recording protocols (SpeechDat)
- Lexicon: ISO-8859 ASCII / SAM-PA (manually corrected)
• SpeechDat orthographic transcription
• Distribution medium: DVD-R

Telephone
The LILA Hindi Belt database collected in India was recorded within the scope of the LILA project. It contains the recordings of 2,023 Hindi speakers (1,011 males and 1,012 females, all speakers with Hindi as first language) recorded over the Indian mobile telephone network. Five dialectal regions have been selected as best representing major Hindi variants in the Hindi belt, focusing on the capital New Delhi. The regions are: Delhi, Northern Uttar Pradesh, Awadh, Rajasthan and central India.

The following acoustic conditions were selected as representative of a mobile user's environment (some speakers were recorded in several environments):
- Passenger in moving car, railway, bus, etc. (311 speakers)
- Public place (300 speakers)
- Stationary pedestrian by road side (300 speakers)
- Home/office environment (712 speakers)
- Passenger in moving car using a hands-free kit (400 speakers)

This database is distributed as 1 Hard Disk. The speech files are stored as sequences of 8-bit, 8kHz A-law speech files and are not compressed, according to the specifications of LILA. Each prompt utterance is stored within a separate file and has an accompanying ASCII SAM label file.

This speech database was pre-validated by SPEX (the Netherlands) to assess its compliance with the LILA format and content specifications.

Each speaker uttered the following items:
- 2 isolated digits
- 2 sequence of 10 isolated digits
- 7 connected digits (1 sheet number -5+ digits, 2 read and 1 spontaneous telephone numbers –9/11 digits, 2 credit card number –14/16 digits, 1 PIN code -6 digits)
- 2 natural numbers
- 2 currency money amounts
- 2 yes/no questions (1 predominantly “yes” question, 1 predominantly “no” question)
- 4 dates (1 spontaneous date e.g. birthday, 2 word style prompted date, 1 relative and general date expression)
- 3 time phrases (1 spontaneous time of day, 2 word style time phrases)
- 6 application words (out of a set of 30)
- 2 spotting phrases using an embedded application word
- 7 directory assistance names (1 spontaneous, e.g. own forename, 1 city of birth/growing up, 2 most frequent cities, 2 most frequent company/agency, 1 “forename surname”)
- 7 spelled words (1 forename, 3 directory assistance city names, 3 real/artificial names for coverage)
- 1 “silence word”
- 6 phonetically rich words
- 15 phonetically rich sentences
- 5 spontaneous items for control
- 10 extra items (Forename Surname, international cities & countries, political parties, VIPs, sports persons, sports teams & horses, ordinals, city names, regions, company names)

The following age distribution has been obtained: 795 speakers are between 16 and 30, 701 speakers are between 31 and 45, and 527 speakers are between 46 and 60.

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