ACE_phonemes

简体中文

updates

• spanish phonemes added

Introduction

We introduce our grapheme-to-phoneme (G2P) conversion dictionary and phoneme lists for the ACE SVS(singing voice synthesis) engine.

Currently, we support three languages: Chinese, Japanese, and English.

Resources

Here is a brief description of each file:

• all_plans.json: this file contains several dictionaries, and the meaning of each dictionary is as follows:
  • jp_word2romaji: Japanese Kana to Romaji dictionary
  • plans: each language has a plan.
    • syllable_alias means each syllable can have multiple spellings.
    • dict contains syllable to phoneme dictionary.
    • phon_class contains all phonemes, head is consonant and tail is vowel
• cmudict.rep: This file contains English pronouncing dictionary, ref: http://www.speech.cs.cmu.edu/cgi-bin/cmudict
• es_final_dict_acesampa_with_seperator.txt: Spanish pronouncing dictionary

G2P (Grapheme-to-Phoneme)

• Chinese and Japanese: Phonemes are directly retrieved from the all_plan.json lookup table.
• English: Phonemes are obtained using a combination of the CMUDict lookup table and G2P model predictions.
• Spanish: Phonemes are predicted using a combination of the lookup table and G2P model.
  • Data is based on IPA-Dict and CharsiuG2P Dict.
  • Minor adjustments have been made to better align with Mexican Spanish pronunciation.
  • For detailed information, refer to /resource/references.

Usage

The main.py file contains basic use cases:

1. Checking whether a phoneme is included in the system.
2. Converting phonemes for Chinese, Japanese, and English.

Contributing

Contributions and suggestions for modification are welcome. You can open an issue or send an email to [email protected].

License

This project is licensed under the MIT License - see the LICENSE file for details.