"This paper deals with the modification of the rule-based duration model by extending the neural network algorithm incorporated into the TTS system Epos. A particular network predicting the sound duration in Czech declarative sentences read in neutral voice is described. The results show that the network was able to learn the characteristics of the speaker. The output generated by the network is much more natural than the previous one, based on simple rules." . "TUDpress" . "2005-09-26+02:00"^^ . "P(1QS108040569), P(GA102/05/0278), Z(AV0Z20670512)" . "Using neural networks to model duration in Czech text-to-speech synthesis" . . "Modelov\u00E1n\u00ED trv\u00E1n\u00ED syntetick\u00E9 \u010De\u0161tiny pomoc\u00ED neuronov\u00FDch s\u00EDt\u00ED"@cs . "Elektronische Sprachsignalverarbeitung" . "Using neural networks to model duration in Czech text-to-speech synthesis" . "Praha" . "RIV/67985882:_____/05:00021878" . "Using neural networks to model duration in Czech text-to-speech synthesis"@en . "Modelov\u00E1n\u00ED trv\u00E1n\u00ED syntetick\u00E9 \u010De\u0161tiny pomoc\u00ED neuronov\u00FDch s\u00EDt\u00ED"@cs . . . . . . . "76;83" . . . "0940-6832" . . "RIV/67985882:_____/05:00021878!RIV08-AV0-67985882" . "8"^^ . . "1"^^ . "Hor\u00E1k, Petr" . . . "Dresden" . "1"^^ . . . . "This paper deals with the modification of the rule-based duration model by extending the neural network algorithm incorporated into the TTS system Epos. A particular network predicting the sound duration in Czech declarative sentences read in neutral voice is described. The results show that the network was able to learn the characteristics of the speaker. The output generated by the network is much more natural than the previous one, based on simple rules."@en . "[A45EF84302E8]" . "speech processing; speech synthesis; neural nets"@en . . "548172" . "Tento p\u0159\u00EDsp\u011Bvek se zab\u00FDv\u00E1 modifikac\u00ED na pravidlech zalo\u017Een\u00E9ho modelu trv\u00E1n\u00ED roz\u0161\u00ED\u0159en\u00EDm syst\u00E9mu neuronov\u00FDch s\u00EDt\u00ED zabudovan\u00FDch v TTS syst\u00E9mu Epos. Je zde pops\u00E1na neuronov\u00E1 s\u00ED\u0165 p\u0159edpov\u00EDdaj\u00EDc\u00ED trv\u00E1n\u00ED hl\u00E1sek v \u010Desk\u00FDch oznamovac\u00EDch v\u011Bt\u00E1ch s neutr\u00E1ln\u00ED prozodi\u00ED. V\u00FDsledky ukazuj\u00ED, \u017Ee neuronov\u00E1 s\u00ED\u0165 je schopna nau\u010Dit se charakteristick\u00E9 rysy mluv\u010D\u00EDho. Trv\u00E1n\u00ED generovan\u00E9 neuronovou s\u00EDt\u00ED je pak p\u0159irozen\u011Bj\u0161\u00ED ne\u017E p\u016Fvodn\u00ED, na jednoduch\u00FDch pravidlech zalo\u017Een\u00FD, model."@cs . . . "Using neural networks to model duration in Czech text-to-speech synthesis"@en .