International Journal of Computer Science Issues

More than a traditional journal...

Efforts to generate speech from written text are different, ranging from mechanical ones of centuries ago, to software solutions of recent decades, known as TTS (Text-to-Speech). The ultimate goal of a speech synthesizer from written text is 'ability' to read any text [2]. Reading should be understandable and natural. A speech synthesizer contains the part for the linguistic analysis (NLP - Natural Language Processing) and the digital signal processing (DSP - Digital Signal Processing)[4]. Albanian language as a separate language in Indo-European family of languages, with its own specifics in writing and reading differs from other languages and models for these languages cannot be used for conversion of Albanian texts. We acquired the Concatenation Synthesis model of building the conversion system of text to speech. This technique is based on preliminary recording of the text, from which acoustic segments are cut and stored in a database. Then, we create application which concatenates these acoustic segments, to create words and sentences [4]. During concatenation of sequences, due to differences in their amplitudes emerge problem of disconnection. In this paper we present some algorithms for minimizing the effect of discontinuity along concatenation of acoustic segments of the synthesized speech [9].