Non-uniform unit selection using fuzzy ARTMAP for memory based Malayalam TTS

Text to speech synthesis system intended for any language, converts the given text in that language to corresponding speech. The major challenge in TTS system is to generate artificial speech which appears to be natural and intelligible. This is essential for visually impaired people to properly understand and comprehend the generated speech. Naturalness can be improved if we mimic the way, human brain generates speech. The memory based model utilizing memory prediction framework is an attempt in this direction. The model learns similar to the way humans learn using Fuzzy ARTMAP. The model employs non-uniform unit selection technique to select appropriate units for synthesizing speech. In this paper, we propose a memory based Malayalam TTS, which uses non-uniform unit selection technique for concatenative speech synthesis.