1.安装 python3.7环境

import os

os.system("rm -rf TensorFlowTTS")

!git clone https://github.com/TensorSpeech/TensorFlowTTS.git

os.chdir("TensorFlowTTS")

!pip install .

os.chdir("..")

import sys

sys.path.append("TensorFlowTTS/")

2.

pip3 install git+https://github.com/repodiac/german_transliterate.git#egg=german_transliterate

pip3 install h5py==2.10.0

3.Load Model

import tensorflow as tf

import yaml

import numpy as np

from tensorflow_tts.inference import AutoConfig

from tensorflow_tts.inference import TFAutoModel

from tensorflow_tts.inference import AutoProcessor

4.Tacotron2

tacotron2 = TFAutoModel.from_pretrained("tensorspeech/tts-tacotron2-baker-ch", name="tacotron2")

FastSpeech2

fastspeech2 = TFAutoModel.from_pretrained("tensorspeech/tts-fastspeech2-baker-ch", name="fastspeech2")

MB-MelGAN

processor = AutoProcessor.from_pretrained("tensorspeech/tts-tacotron2-baker-ch")

input_ids = processor.text_to_sequence("这是一个开源的端到端中文语音合成系统,我喜欢你哦", inference=True)
 

_, mel_outputs, stop_token_prediction, alignment_history = tacotron2.inference(   tf.expand_dims(tf.convert_to_tensor(input_ids, dtype=tf.int32), 0),  tf.convert_to_tensor([len(input_ids)], tf.int32),   tf.convert_to_tensor([0], dtype=tf.int32)  )

remove_end = 1024
audio = mb_melgan.inference(mel_outputs)[0, :-remove_end, 0]

import soundfile
soundfile.write('test.wav',audio.numpy(),24000)

播放文件ok

参考https://colab.research.google.com/drive/1YpSHRBRPBI7cnTkQn1UcVTWEQVbsUm1S?usp=sharing#scrollTo=6-RiC7nafHQk

以下也有demo

https://huggingface.co/tensorspeech/tts-fastspeech2-baker-ch

https://huggingface.co/tensorspeech/tts-tacotron2-baker-ch

可惜官方没有测试的asr

可以简单用用