ESCA Tutorial and Research Workshop on Speech Input/Output Assessment and Speech Databases Noordwijkerhout, The Netherlands September 20-23, 1989

The Speech Database Used in SPICOS

Volker Steinbiss, Hans-Hermann Hamer, Dieter Mergel, Hermann Ney, Andreas Noll, Annedore Paeseler, Herbert Piotrowski, Horst Tomaschewski

Philips GmbH Forschungslaboratorium Hamburg, Hamburg, F.R. Germany

The speech database recorded for the speech recognition and understanding system SPICOS is described.

The 100 "Berlin sentences" are representative of the phoneme distribution of the German language and are used to train the speech recognition module. Recognition tests are performed on the basis of the 200 so-called "SPICOS test sentences". These sentences are typical for the SPICOS application (database queries) and are fairly different from the Berlin sentences: They are longer and cover a different vocabulary.

Recordings were taken of two female and three male speakers, with each of these covering two sessions of the Berlin sentences and three sessions of the SPICOS sentences, each fluently spoken. For some of the speakers, there are also recordings of words spoken in isolation: the 917 SPICOS words and the vocabulary of the Berlin sentences. All recordings were made using a close-talking microphone and simultaneously via telephone line.

Full Paper

Bibliographic reference.  Steinbiss, Volker / Hamer, Hans-Hermann / Mergel, Dieter / Ney, Hermann / Noll, Andreas / Paeseler, Annedore / Piotrowski, Herbert / Tomaschewski, Horst (1989): "The speech database used in SPICOS", In SIOA-1989, Vol.2, 59-62.