Deployment solutions for business users.

Hardware: #

The hardware configuration depends mainly on the local audio inference hourly requirements, with no hard and fast performance requirements.

Requires nvidia gpu.

minimum configuration #

• CPU: intel 13th generation i5 or amd equivalent and above models
• Memory: 16G
• GPU: NVIDIA graphics card with 8G of video memory and above
• Hard disk: 50G or more

Recommended Configurations #

• CPU: intel 13th generation i7 or amd same level and above models
• Memory: 32G
• GPU: NVIDIA graphics card with 24G RAM and above
• Hard disk: 100G or more

Note: We tested in a configuration of 13700K + 64G RAM + A6000 GPU (arithmetic equivalent to NVDIA RTX3090), which can reasonably generate about 25-50 seconds (GPU time) of animation per second. (That is, the current configuration theoretically supports 25-50 users with 0-latency concurrency, but it is only a theoretical value, and difficult to do in practice.)

---

System: #

docker is a cross-platform deployment tool that can be deployed on all docker-enabled platforms (linux, mac, win), but the release version of the AnimaCore docker image has only been tested on Ubuntu 22.04 LTS and Windows 10, and other platforms need to be tested on their own. Please feel free to submit suggestions for other platforms and problems you encountered on other platforms.

Ubuntu. #

Environment:

• Graphics Card Driver
• docker(docker compose)
• nvidia-smi
• NVCC
• nvidia-docker2
• python:(comes with the system)

Windows. #

Environment:

• Install docker desktop
• Install python 3.10 (install pyyaml: pip install pyyaml)

---

Deployment: #

1. Run load_images.sh to load docker image. (Right click load_images.ps1 under window to run)

cd /docker_images
. /load_images.sh

2. Unzip ServerRun_Local.zip

3. Catalog structure:

MHCtalker_localServer_0.x.x/

├── docker_images/

│ ├── main_l_00x.tar

│ ├── asr_00x.tar

│ └── ......

│

└── ServerRun_Local/

│ ├── main

│ ├── run.pyc

│ └── ......

│

4. Activation

cd /ServerRun_Local
sudo python run.pyc

Windows

The animation module starts slowly, the whole startup process takes about 5 minutes

---

Database: #

The local database is a sqlite database located at /ServerRun_Local/main/db/database.db

The database currently has no encryption settings and permission settings, the next version may add permission settings or replace the mysql database to ensure data security.

---

Client Connection: #

The local server is running on port 19875 of the local machine, please make sure the port is not occupied by other programs.

Client connection (UE5-Project Settings-MHC_Talker-url) address: ws://:19875

---

Other parameters: #

See /ServerRun_Local/main/.env

AUTH_KEY	lic_key
GREETING	Whether or not to return the welcome prompt statement when connecting.

Fill in the following parameters as needed, the qianqi api when you need to enable the conversation function, azure api when you need the voice function.

If only the animation module is used, leave the following parameters blank

QWEN_API_BASE	Tongyi Qianqian api base
QWEN_API_KEY	tonyiqianqian api key
–	–
AZURE_SPEECH_BASE	Microsoft Azure Voice base
AZURE_SPEECH_KEY	Microsoft Azure Voice Key
AZURE_SPEECH_REGION	Microsoft Azure Voice Regions

---