2511

2761پ

2674

كارشناسي ارشد

مخابرات

اصفهان: دانشگاه صنعتي اصفهان، دانشكده برق و كامپيوتر

1382

[هشت ]، 77ص .: مصور، جدول، نمودار

فرامرز هندسي

عليمحمد دوست حسيني

23/03/84

1396/10/25

كتابنامه

برق و كامپيوتر

مهندسي برق و كامپيوتر

ID2674

به فارسي و انگليسي : قابل رويت در قسمت " نسخه ديجيتال "

AbstractThe accuracyof text analysis in text to speechconversionsystems a major effect in the quality hasof the synthesizedspeech There are severalways in which Persianwords can acquire phoneticdescription The first and most commonmethodis to assignphonemes eachword and storethese toassignments a lexicon In this technique a da tabase searched find the vowels for eachword in is toThis is known as the dictionary method The dictionary method is accurate but it needsa hugedatabasethat consumessignificant resources and accesstime for such a databasewould beconsiderable Another approach usesa neural network NN to extractthe structureof the Persianvowelization For this purpose training data should be gathered that is representativeof the language Using a neural network for phonememappingin text to speech conversioncreatesa fast distributedrecognitionenginethat supportsthe mappingof missingwords in the database Becauseof the complexityof the vowelizationprocessin Persian using a NN to extract all of the requiredstructures leads to unacceptableaccuracy So a combination of the Neural Networks and aprobabilistic methodin additionto utilizing somerules in Persianlanguage a reasonablesolution isfor achieving high performance At last the probabilistic model is evaluatedalone in the Persiantext analyzer It is shown that by choosing proper numberof parametersand training sequence high performancecanbe achieved I of 0 A t Wla Iifi Lf V IJ W

فرامرز هندسي

عليمحمد دوست حسيني