Neural talk: Text-to-speech synthesizer using neural networks

Neural Talk is a system developed to convert written text into audible speech. This PC-based system employs the use of a neural networks in converting the words to phonemes. Synthesized speech is produced using the Sound Blaster medium. The neural networks implemented is the back-propagation as its training algorithm. The project uses its own set of phonemes for compatibility with the networks\u27 input parameters. The system achieves a high accuracy rate through a network that has been developed to learn the correct assignment of phonemes to a word. The input and output coding of the 1,000 word data contains an innate knowledge for the neural network to learn. The training emphasizes primarily on the networks\u27 capability to learn word intonations through the phonemes. The output phonetic word interpretations are used to synthesize the speech. The system uses a 7-letter window based on NETtalk technology in the learning process. The networks gain better performance in generalizing new data through the process of learning more words. The training speed increases when network size is reduced . It learns consonants faster than vowels. The system\u27s characteristics are similar to those of phoneme structure. Basically, the main advantage of this system is its capability to convert the text into speech without the need for a word library as a reference, unlike in other applications that make use of the artificial intelligence technique