/lec-chatbot

Primary LanguagePython

Langex chatbot for detecting pronunciation accuracy using STT model

1. Todos:

[ ] Understand how ctranslate can optimize inference process

2. Deploy:

Whisper dependencies

  • ffmpeg CLI
sudo apt update && sudo apt install ffmpeg
  • OpenAI's tiktoken => Use Rust if tiktoken does not work.
  • ffmpeg-python

2. Extension to xmpp msg

2.1. Message extension

<message>
	<body>
	</body>
	<langexbot xmlns="langex:chatbot">
		<feature-element>
	</langexbot>
</message>
  1. Onboard
	<onboard xmlns="langex:chatbot:onboard"> </onboard>
  1. Pronunciation Assessment
  <pronunc_assess xmlns="langex:chatbot:pronunc_assess"> </pronunc_assess>

3. Docker

3.1. Build image

To effectively build the image by not reinstalling all packages whenever requirements.txt changed. You should execute sudo make config-dockerd to config daemon.json for docker daemon. Be careful that command will override all current configurations daemon.json. If it is in case, you should manually append configuration in daemon.json file in the current working directory to that system configuration daemon.json file.

3.2 Run container

Model is installed at /root/.cache/huggingface and /usr/local/src/lec-chatbot/chatbot/models/whisper/model. Please mount it to avoid downloading model whenever run the container.

docker run -d --rm narutosimaha/lec-chatbot

4. Deployment backlog

[ ] Set up certificate for file service.