Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

The VESTEL Telephone Speech Database

D. Tapias, Alejandro Acero, J. Esteve, J. C. Torrecilla

Telefonica Investigation y Desarrollo, Madrid, Spain

VESTEL is a telephone speech corpus collected at the Speech Technology Division of Telefonica Investigacidn y Desarrollo. The data base was designed to support research in speaker-independent automatic speech recognition (ASR) based on word and subword units. Over sixteen thousand people called in response to newspaper advertisements. They were prompted by a recorded voice to say digits, numbers and commands, and to answer questions asking them the city where they lived and they were born, their name and surnames, a yes/no question and to spell some words. The utterances were spoken over commercial telephone lines, and each call was composed by twenty five separate utterances. Spain was divided into ten dialectical regions in order to take into account the main Spanish dialects of Castilian (usually known as "Spanish"). Each call was checked and transcribed by two people. In this report we describe the system implemented to record the data base, the publicity campaign, the recording protocol, the regions in which Spain was divided, and the statistical information of the tasks that were carried out.

Full Paper

Bibliographic reference.  Tapias, D. / Acero, Alejandro / Esteve, J. / Torrecilla, J. C. (1994): "The VESTEL telephone speech database", In ICSLP-1994, 1811-1814.