AUTOMATIC SPEECH RECOGNITION USING RASPBERRY PI PLATFORM

U ovom radu obrađena je osnovna podjela sustava za automatsko prepoznavanje govora koji su TTS (Text-to-speech) i STT (Speech-to-text) te njegova povijest. Objašnjene su glavne značajke govora koje se mora uzeti u obzir prilikom projektiranja takvih sustava. Nadalje, objašnjeni su algoritmi i alati za prepoznavanje govora kao što su dinamičko savijanje vremena, umjetne neuronske mreže, skriveni Markovljevi modeli te dubinske neuronske mreže. Također objašnjena je arhitektura sustava za sintetiziranje govora te su opisane sve analize koje sustav odrađuje prilikom sintetiziranja. U posljednjem dijelu rada opisana je implementacija aplikacije za prepoznavanje govora (Voicecommand) unutar Raspbian operativnog sustava.This thesis consists of describing core segmentation of automatic speech recognition systems, which is TTS (Text-to-speech) and STT (Speech-to-text), and it's history. Most important word features which have to be taken into account during projecting those systems are explained. Furthermore, algorithms and tools used in speech recognition are also described, some of them being dynamic time-warping, artificial neural networks, hidden Markov models and deep neural networks. Architecture of systems for speech synthesizing is also described with all analysis needed for speech synthesizing. Last part of thesis consists of describing the implementation of one of many speech recognition applications (Voicecommand) inside of Raspbian operating system