Storing prosody attributes of spontaneous speech

The paper deals with the problem of the storage of the prosody attributes which have been used in the Czech dialog system. In Czech language featured by a free-word-ordering the prosody serves a critical information for the recognition and understanding system. For some sentences the intonation is essential to determine the core of a communication, beeing used by a speaker who to emphasize a meaning of a sentence. The prosodic characteristics included in the sentence (features describing fundamental frequency F0, voice energy, the length of a pause behind and before the word, the speaking rate, flags indicating word finality and the lexical word accent) are stored in the database and consequently exploited by the linguistic module as an additional information used for recognizing and understanding the spontaneous speech. In case of storing digital speech signal in database we meet a problem of its high redundancy. Another problem is choosing cut points for segmenting speech. Suitable points are pauses, but they are not often present in the fluent speech. Instead a coarticulation between phonemes makes the placement of the cut points difficult. Their suitably for segmenting speech should be dependent on a context information. Having processed the characteristics by usual methods of statistics the database can also be used to generate answers in the dialog system. The module was implemented in the C language and supported by the ORACLE database.