Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

The data collection of Voice Across Japan (VAJ) project

Ikuo Kudo, Takao Nakama, Nozomi Arai, Nahoko Fujimura

Texas Instruments, Tsukuba Research and Development Center Ltd., Ibaraki, Japan

The Voice Across Japan (VAJ) speech data collection started from September, 1993, and is now in progress. VAJ is a large-scale telephone database to support the creation of continuous, speaker-independent speech recognition systems. The goal of the effort is to obtain speech from 10,000 speakers with broad dialect coverage, then use the database for developing recognition systems ranging from digit recognition system to large vocabulary systems based on phonetic recognition units. In collecting a 10,000 speaker corpus in all Japan through telephone line, both data quality and data sampling become very important problems. In order to improve the problem, this paper describe our method of data collection and reports the current status of our database such as balanced data set. keyword Speech Database, Voice Across Japan (VAJ), Telephone speech, Speaker-independent speech recognition, robustness, Digit recognition, Tri-phone.

Full Paper

Bibliographic reference.  Kudo, Ikuo / Nakama, Takao / Arai, Nozomi / Fujimura, Nahoko (1994): "The data collection of voice across Japan (VAJ) project", In ICSLP-1994, 1799-1802.