中文

A demo of zh/Chinese Text to Speech system run on CPU in real time. (fastspeech2 + mbmelgan)

RTF(real time factor): 0.2 with cpu: Intel(R) Core(TM) i5-7200U CPU @ 2.50GHz 24khz audio use fastspeech2, RTF1.6 for tacotron2

This repo is mainly based on TensorFlowTTS with little improvement.

• tflite model come from colab, thx to @azraelkuan
• add pause at punctuation
• add TN (Text Normalization) from chinese_text_normalization

text = "2020年，这是一个开源的端到端中文语音合成系统"

zhtts synthesis mp3

pip install zhtts

or clone this repo, then pip install .

import zhtts

text = "2020年，这是一个开源的端到端中文语音合成系统"
tts = zhtts.TTS() # use fastspeech2 by default

tts.text2wav(text, "demo.wav")
>>> Save wav to demo.wav

tts.frontend(text)
>>> ('二零二零年，这是一个开源的端到端中文语音合成系统', 'sil ^ er4 #0 l ing2 #0 ^ er4 #0 l ing2 #0 n ian2 #0 #3 zh e4 #0 sh iii4 #0 ^ i2 #0 g e4 #0 k ai1 #0 ^ van2 #0 d e5 #0 d uan1 #0 d ao4 #0 d uan1 #0 zh ong1 #0 ^ uen2 #0 ^ v3 #0 ^ in1 #0 h e2 #0 ch eng2 #0 x i4 #0 t ong3 sil')

tts.synthesis(text)
>>> array([0., 0., 0., ..., 0., 0., 0.], dtype=float32)

clone this repo, pip install flask first, then

python app.py

• visit http://localhost:5000 for tts interaction
• do HTTP GET at http://localhost:5000/api/tts?text=your%20sentence to get WAV audio back:

$ curl -o "helloworld.wav" "http://localhost:5000/api/tts?text=%E4%BD%A0%E5%A5%BD%E4%B8%96%E7%95%8C"

%E4%BD%A0%E5%A5%BD%E4%B8%96%E7%95%8C is url code of"你好，世界！"

wav generate from tacotron model is better than fast speech, however tacotron is much slower , to use Tacotron, change code

import zhtts
tts = zhtts.TTS(text2mel_name="TACOTRON")
# tts = zhtts.TTS(text2mel_name="FASTSPEECH2")