Naozumi0512/g2p-Cantonese-aggregate

g2p Dataset - Cantonese (yue)

Description

The Cantonese (yue) language corpus for the G2P (Grapheme-to-Phoneme) task consists of 68,500 Cantonese words/phrases, along with corresponding phonetic transcriptions for randomly selected polyphonic characters.

This dataset is formatted to align with the CPP (Chinese Polyphones with Pinyin) structure.

Languages

• Cantonese (yue)

File Descriptions

*.sent: Contains words/sentences, with one polyphonic character wrapped by an ANCHOR_CHAR (▁) [Not underscore] per line. Only one polyphonic character is marked per sentence, distributed randomly for maximum coverage and even distribution.
*.lb: Matches the .sent file, with each line containing the Jyutping transcription of the marked polyphonic character in .sent.
*.pos: Matches the .sent file, with each line containing the POS tag of the marked polyphonic character in .sent, generated using g2pW-Cantonese/scripts/tag_pos.py. (See original project)

Limitations:

Due to the collected data type, each line in the dataset may be too short (mostly words and phrases, rarely full sentences), which may limit context-based performance. I have created also a extended dataset on top of this: https://huggingface.co/datasets/Naozumi0512/g2p-Cantonese-aggregate-extended

Data Source

• Rime Cantonese Input Schema (jyut6ping3.words.dict.yaml)
• 粵典 Words.hk
• CantoDict

Original Project

https://github.com/Naozumi520/g2pW-Cantonese
https://huggingface.co/Naozumi0512/g2pW-Cantonese

License

Inherits licenses from source materials (CC-BY-4.0 and others).