基于Python PaddleSpeech实现语音文字处理

前言

这段时间一直在研究飞浆平台，最近试了试PaddleSpeech项目，试着对文本语音做处理。整体的效果个人觉着不算特别优越，只能作为简单的学习使用。

项目github地址：github仓库

环境安装

首先我们看一下项目结构以及安装文档。

需要Python3.7以上、C++环境、requirements安装等等，下面按照我的顺序说一下。

1、conda安装Python3.9虚拟环境

使用conda安装python3.9环境，命令如下。

conda create -n py39 python=3.9

2、安装Visual Studio 2019

安装地址: Microsoft C++ 生成工具 - Visual Studio

注意安装的时候需要勾选C++桌面开发。

3、安装requirements.txt

使用命令安装requiremets.txt，命令如下：

pip install -r requirements.txt -i https://pypi.douban.com/simple

这里要注意一下，paddlespeech_ctcdecoders安装失败的话无所谓，可以略掉。

4、安装paddlepaddle和paddlespeech

命令如下：

pip install paddlepaddle -i https://mirror.baidu.com/pypi/simple
pip install paddlespeech -i https://pypi.tuna.tsinghua.edu.cn/simple

5、nltk_data下载

按照项目安装文档内的说明。

我的本地目录地址如下

项目验证

我下面分别验证一下tts、asr以及标点恢复功能。

tts语音合成

使用命令如下：

paddlespeech tts --input "南京现在很冷，下次再去夫子庙吧。" --output C:\Users\xxx\Desktop\115.wav

执行过程

(dh_partner) D:\spyder\PaddleSpeech>paddlespeech tts --input "南京现在很冷，下次再去夫子庙吧。" --output C:\Users\xxx\Desktop\115.wav
phones_dict: None
[2022-01-05 17:23:43,642] [    INFO] [log.py] [L57] - File C:\Users\huyi\.paddlespeech\models\fastspeech2_csmsc-zh\fastspeech2_nosil_baker_ckpt_0.4.zip md5 checking...
[2022-01-05 17:23:44,742] [    INFO] [log.py] [L57] - Use pretrained model stored in: C:\Users\huyi\.paddlespeech\models\fastspeech2_csmsc-zh\fastspeech2_nosil_baker_ckpt_0.4
self.phones_dict: C:\Users\huyi\.paddlespeech\models\fastspeech2_csmsc-zh\fastspeech2_nosil_baker_ckpt_0.4\phone_id_map.txt
[2022-01-05 17:23:44,743] [    INFO] [log.py] [L57] - C:\Users\huyi\.paddlespeech\models\fastspeech2_csmsc-zh\fastspeech2_nosil_baker_ckpt_0.4
[2022-01-05 17:23:44,744] [    INFO] [log.py] [L57] - C:\Users\huyi\.paddlespeech\models\fastspeech2_csmsc-zh\fastspeech2_nosil_baker_ckpt_0.4\default.yaml
[2022-01-05 17:23:44,744] [    INFO] [log.py] [L57] - C:\Users\huyi\.paddlespeech\models\fastspeech2_csmsc-zh\fastspeech2_nosil_baker_ckpt_0.4\snapshot_iter_76000.pdz
self.phones_dict: C:\Users\huyi\.paddlespeech\models\fastspeech2_csmsc-zh\fastspeech2_nosil_baker_ckpt_0.4\phone_id_map.txt
[2022-01-05 17:23:44,745] [    INFO] [log.py] [L57] - File C:\Users\huyi\.paddlespeech\models\pwgan_csmsc-zh\pwg_baker_ckpt_0.4.zip md5 checking...
[2022-01-05 17:23:44,782] [    INFO] [log.py] [L57] - Use pretrained model stored in: C:\Users\huyi\.paddlespeech\models\pwgan_csmsc-zh\pwg_baker_ckpt_0.4
[2022-01-05 17:23:44,783] [    INFO] [log.py] [L57] - C:\Users\huyi\.paddlespeech\models\pwgan_csmsc-zh\pwg_baker_ckpt_0.4
[2022-01-05 17:23:44,783] [    INFO] [log.py] [L57] - C:\Users\huyi\.paddlespeech\models\pwgan_csmsc-zh\pwg_baker_ckpt_0.4\pwg_default.yaml
[2022-01-05 17:23:44,785] [    INFO] [log.py] [L57] - C:\Users\huyi\.paddlespeech\models\pwgan_csmsc-zh\pwg_baker_ckpt_0.4\pwg_snapshot_iter_400000.pdz
vocab_size: 268
frontend done!
encoder_type is transformer
decoder_type is transformer
C:\Users\huyi\.conda\envs\dh_partner\lib\site-packages\paddle\framework\io.py:415: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' i
s deprecated since Python 3.3, and in 3.10 it will stop working
 if isinstance(obj, collections.Iterable) and not isinstance(obj, (
acoustic model done!
voc done!
Building prefix dict from the default dictionary ...
[2022-01-05 17:23:51] [DEBUG] [__init__.py:113] Building prefix dict from the default dictionary ...
Loading model from cache C:\Users\huyi\AppData\Local\Temp\jieba.cache
[2022-01-05 17:23:51] [DEBUG] [__init__.py:132] Loading model from cache C:\Users\huyi\AppData\Local\Temp\jieba.cache
Loading model cost 0.659 seconds.
[2022-01-05 17:23:52] [DEBUG] [__init__.py:164] Loading model cost 0.659 seconds.
Prefix dict has been built successfully.
[2022-01-05 17:23:52] [DEBUG] [__init__.py:166] Prefix dict has been built successfully.
C:\Users\huyi\.conda\envs\dh_partner\lib\site-packages\paddle\fluid\dygraph\math_op_patch.py:251: UserWarning: The dtype of left and right variables are not the same, left dtype is padd
le.int64, but right dtype is paddle.int32, the right dtype will convert to paddle.int64
 warnings.warn(
[2022-01-05 17:23:58,811] [    INFO] [log.py] [L57] - Wave file has been generated: C:\Users\xxx\Desktop\115.wav

生成的音频如下

asr语音识别

我就使用了tts生成的音频进行asr识别，看看效果，命令如下:

paddlespeech asr --lang zh --input C:\Users\xxx\Desktop\115.wav

执行结果如下

可以看到最后打印的内容是没有标点的文字输出，还是比较准的。

标点恢复

就用这句话试试标点恢复的情况，命令如下：

paddlespeech text --task punc --input 南京现在很冷下次再去夫子庙吧

执行结果

看起来语义上没什么问题。

来源：https://blog.csdn.net/zhiweihongyan1/article/details/122326644