Nepali ARPABET Phonetic Dictionary

This repository contains a comprehensive ARPABET-based phonetic dictionary for the Nepali language, designed to support text-to-speech (TTS) systems in low-resource settings. ARPABET, adapted from English phonetics, provides ASCII-compatible representations for Nepali's Devanagari script, including nuances like aspiration, nasalization, and stress. This resource was developed as part of an undergraduate project at Tribhuvan University's Pashchimanchal Campus, focusing on neural TTS architectures like Tacotron2 and HiFi-GAN.

Files

- 3000Merged.txt: A merged dictionary with over 3,000 entries, combining automated generation and manual refinements for broad coverage of common Nepali words and phoneme mappings.- Manually Curated.txt: Hand-curated entries emphasizing accuracy for high-frequency terms, dialectal variations, and irregular pronunciations.- Rule Based.txt: Rule-based mappings for grapheme-to-phoneme conversion, handling Devanagari-specific rules like conjunct consonants and vowel diacritics.

Usage

These files follow the CMUdict format: `WORD  PHONEME1 PHONEME2 ... PHONEMEn`. Integrate them into G2P pipelines or TTS models for phonetic transcription. Example:

नेपाली N EH0 P AA1 L IY0

Credits

Created by Aayush Adhikari, Shishir Rijal, Roshan Tiwari, and Sudip Acharya, under the supervision of Nabin Lamichhane (Department of Electronics and Computer Engineering, Pashchimanchal Campus, Pokhara, Nepal).

For contributions or issues, please open a pull request or contact the authors.