利用speechbrain和openai分别实现语音和文字的情绪识别
环境需求
硬件
Nvidia GPU Drive 50G Memory 16G
系统
Ubuntu 20.04 LTS Server CUDA 11.4
请确认已安装好CUDA驱动，可以用下面这个命令
ubuntu@VM-0-12-ubuntu:~$ nvidia-smi
Sat May 6 09:13:09 2023
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.82.01 Driver Version: 470.82.01 CUDA Version: 11.4 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 Tesla T4 On | 00000000:00:08.0 Off | Off |
| N/A 27C P8 9W / 70W | 0MiB / 16127MiB | 0% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
如果没有出现以上结果请去NVIDIA官网下载并安装对应版本的驱动，并按照指引安装。注意，尽量不要使用带有GUI的ubuntu，有一定几率会出现驱动冲突从而无法正常显示。 CUDA网址：https://developer.nvidia.com/cuda-toolkit-archive
配置过程
1、安装Conda
由于是服务端，所以安装轻量版无GUI的MiniConda
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh
Do you accept the license terms? [yes|no]
[no] >>> yes
Do you wish the installer to initialize Miniconda3
by running conda init? [yes|no]
[no] >>> yes
# 其他选项默认，不用sudo
安装完成后
ubuntu@VM-0-12-ubuntu:~$ source .bashrc
(base) ubuntu@VM-0-12-ubuntu:~$
出现（base）代表安装成功
2、准备代码与模型文件，准备conda环境
下载 speechbrain-restful.tar 文件，并上传至～/目录
解压 speechbrain-restful.tar 到 ～/
(base) ubuntu@VM-0-12-ubuntu:~$ tar -xvf speechbrain-restful.tar
(base) ubuntu@VM-0-12-ubuntu:~$ ls speechbrain/
audio_cache docs InferText.py lint-requirements.txt pretrained_models pyproject.toml README.md requirements.txt results speechbrain tests
conftest.py InferSpeech.py LICENSE pretrained __pycache__ pytest.ini recipes restfulTest.py setup.py templates tools
3、准备Conda环境
下载 emotion_conda23.3.1.tar.gz 文件，并上传至～/目录
解压 emotion_conda23.3.1.tar.gz
(base) ubuntu@VM-0-12-ubuntu:~$ mkdir emotion
(base) ubuntu@VM-0-12-ubuntu:~$ mv emotion_conda23.3.1.tar.gz emotion
(base) ubuntu@VM-0-12-ubuntu:~$ cd emotion
(base) ubuntu@VM-0-12-ubuntu:~/emotion$ tar -zxvf emotion_conda23.3.1.tar.gz
将 ～/emotion 移动到 ～/miniconda3/envs/
(base) ubuntu@VM-0-12-ubuntu:~$ mv emotion miniconda3/envs/
(base) ubuntu@VM-0-12-ubuntu:~$ cd miniconda3/envs/
(base) ubuntu@VM-0-12-ubuntu:~/miniconda3/envs$ ls
emotion
激活环境，确认环境迁移正确
(base) ubuntu@VM-0-12-ubuntu:~$ source miniconda3/envs/emotion/bin/activate
(emotion) ubuntu@VM-0-12-ubuntu:~$ conda-unpack
(emotion) ubuntu@VM-0-12-ubuntu:~$ conda deactivate
(base) ubuntu@VM-0-12-ubuntu:~$ conda env list
# conda environments:
#
base * /home/ubuntu/miniconda3
emotion /home/ubuntu/miniconda3/envs/emotion
至此，所有内容都已准备完成
4、测试
要记得修改InferText.py中的url地址，这个url是使用 https://github.com/cookeem/chatgpt-service 搭建的openai服务，参考这个仓库的教程搭建好服务，修改ip地址即可。
(base) ubuntu@VM-0-12-ubuntu:~$ conda activate emotion
(emotion) ubuntu@VM-0-12-ubuntu:~$ cd speechbrain/
(emotion) ubuntu@VM-0-12-ubuntu:~/speechbrain$ ls
audio_cache docs InferText.py lint-requirements.txt pretrained_models pyproject.toml README.md requirements.txt results speechbrain tests
conftest.py InferSpeech.py LICENSE pretrained __pycache__ pytest.ini recipes restfulTest.py setup.py templates tools
(emotion) ubuntu@VM-0-12-ubuntu:~/speechbrain$ python restfulTest.py
* Serving Flask app 'restfulTest'
* Debug mode: on
WARNING: This is a development server. Do not use it in a production deployment. Use a production WSGI server instead.
* Running on all addresses (0.0.0.0)
* Running on http://127.0.0.1:5000
* Running on http://172.26.0.12:5000
Press CTRL+C to quit
* Restarting with stat
* Debugger is active!
* Debugger PIN: 106-610-369
使用postman测试均正常
