Multi-Module G2P Converter for Persian Focusing on Relations between Words

G2P systems aim to convert a grapheme (letter) sequence into its In this paper, we investigate the application of pronunciation sequence, and are an essential end-to-end and multi-module frameworks for component of text-to-speech (TTS) and speech G2P conversion for the Persian language. The recognition systems for any language lacking results demonstrate that our proposed multimodule consistent pronunciation rules. G2P system outperforms our end-to-end A good G2P system must address the issues of systems in terms of accuracy and speed. The out-of-vocabulary (OOV) words and cross-word system consists of a pronunciation dictionary as relations. OOV words are those which are not our look-up table, along with separate models to present in the lexicon, meaning they were not handle homographs, OOVs and ezafe in Persian seen during model training. In the case of G2P, created using GRU and Transformer the lexicon is a dictionary consisting of architectures. The system is sequence-level rather graphemes and their respective phonemes. As for than word-level, which allows it to effectively cross-word relations, a Persian G2P task is capture the unwritten relations between words mainly concerned with homographs and ezafe (cross-word information) necessary for constructions.