Chin-concatTTS

KO | EN

gitlite — search

#python #react #ai #llm #java #nodejs #cli #kubernetes #awesome-list #vue #mcp #openai #ios #open-source

(★ 9)

Cantonese and Mandarin concatenation TTS system in Python

Chin-concatTTS 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Chin-concatTTS
Cantonese and Mandarin concatenation TTS system in Python 
For DNN TTS, please check https://github.com/kennethli319/Chin-dnnTTS

# Usage
 
New users please install the following external modules: 
 pip install -U pycantonese 
 pip install opencc-python-reimplemented 
 pip install pkuseg

Usage (developed in python3): 
 python word_syn.py <"input sequence"> <-language c or p> <-play> <-volume 0-100> <-crossfade> <-outfile filename>

Example: 
 python3 word_syn.py "1/1/2001，999！翻译都要执行多个翻译系统，这带来巨大的计算成本。如今，许多领域都正在被神经网路技术颠覆。" -l p -p -v 80 -c -o output_mandarin.wav 
 <a href="https://drive.google.com/open?id=16t2mE66eJdZEL4jB1_h01Q__zcg5vUax"> output_mandarin.wav </a> 
 python3 word_syn.py "1/01/1991，32。翻译都要执行多个翻译系统，这带来巨大的计算成本。如今，许多领域都正在被神经网路技术颠覆。" -l c -p -v 80 -c -o output_cantonese.wav 
 <a href="https://drive.google.com/open?id=10DRGh6Lf3ABBM9Kj1bSCjM2qj7sjRhr6"> output_cantonese.wav </a> 
 
# LOGBK and PROBLEMS
16 DEC - Done word-wav data in Can and Manderin 
18 DEC - Done overall documentation 
19 DEC - (ING) Class structure -> token -> multichar prob! 
19 DEC - NSW Date conversion 
23 DEC - Done crossfade 
23 DEC - Updated cantonese phonedict (according to website and hkcan corpus per percentage) 
29 DEC - Updated all README and docString

1 - [DONE-23dec] check sim chin / tran chin and convert (removed checking part) 
2 - [DONE-23dec] cantonese corpus = very slow -> change it to a txt file with count / % only 
3 - [DONE-18dec] change chin char (simplified chinese vs traditional chinese) 
4 - [DONE-18dec] word seg: use pku module, but not yet merge to the structure 
5 - [Start-29dec] add POS tag 
5 - [AfterPOS-?] multi pronouncaition for the same word -> depends on the previous history : can be solve by POS + history(or word seg token)

# Pipeline
DATA STRUCTURE 
 CLASS:
 Sequence: sequence 
 Char: char info 
 i) Char 
 ii) CV - onset, n, coda 
 iii) Tone 
PROCEDURE

[PREPARE] 
0 - Import necessary libraries 
1 - Argv management and declear global variables 
2 - Define Methods and Classes 
 2.1 - Methods 
 2.2 - Classes

[MAIN] 
3 - Frontend : 
 3.1 - Get sequence 
 3.2 - Normalization -> remove (unnecessary) punct, change all remaining punct to 全/半形 
 3.3 - Classify cantonese / mandarin / english(not yet) / mixture 
 3.4 - Word seg -> to tokens 
 3.5 - POS of token seq 
 3.6 - Based on POS -> check the phones/words/diphones (from dict) 
 3.7 - Change the phone/diphone/word-pron seq according to Consonant-Vowel Coarticulation rules in that lang 
 3.8 - Intonation 
4 - Waveform Generation 
 4.1 - WORD UNIT VERION 
 4.1.1 - load all required word wav 
 4.1.2 - smoothing on boundaries? 
 4.1.3 - output fine tune: eg volume/speed 1x 1.5x 2x? 
 4.2 - DIPHONE UNIT VERSION 
5 - Output 
 5.1 - Output wav 
 5.2 - Output all relations 
 5.3 - Output all parameters 
 5.4 - Output full info

Chin-concatTTS

Chin-concatTTS 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Chin-concatTTS

Chin-concatTTS 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README