Interview robot based on ChatGPT

ChatGPT acts as the interviewer / interviewee. It will reponse to you with realistic voice. Moreover, it can become your shadow that means you can transfer your personal info to it making it more akin to you.

Pipeline

• speech2text function based on Whisper API

# you can create test audio track via:
https://online-voice-recorder.com/

• text2speech function

# you can transform text to speech via:
https://elevenlabs.io/

• ChatGPT uses chat history stored in your DB (MongDB or Json)

Architecture

Project environment setup and test

• pip install

pip install -r requirements.txt

• create file .env (get from https://platform.openai.com/)

OPEN_AI_ORG=
OPEN_AI_KEY=

• GUI (fastAPI)

# start/reload the main file
uvicorn main:app --reload
# or 
http://127.0.0.1:8000/docs#/

Improvement

• background noise of recording can be removed

Install ffmpeg from ffmpeg.org
Version: ffmpeg version 2024-01-04-git-33698ef891-essentials_build-www.gyan.dev

Implement spectral gating noise reduction algorithm
https://github.com/timsainb/noisereduce

Progress

Item	Progress
speech2txt	✅
gpt response	✅
personal shadow creation	✅
txt2speech	🕑
front-end gui	✅
various speech input	✅
various speech voice output	🕑