A voice2voice chatgpt assistant
Constructed by using OpenAI Whisper + OpenAI ChatGPT API + Google Text2Speech Service

• Speech2Text through OpenAI's Whisper Model (currently using local CPU)
• Chat with ChatGPT through its API
• Text2Speech through Google's Text2Speech Service
• Related tools
  • sox: play the .mp3 files
  • arecord: record your voices through microphone (ubuntu default toolset)
  • lame: transform arecord's raw data to .mp3 file

• We can now ask ChatGPT to reset the session for us. Therefore it will clear out the current session, preventing spend the quota on unrelated history messages.

• Whisper would automatically download model for the first time
• Make sure use a python virtual env before start
• Currently, only 1 background session available at any time

Run the following command manually or using scripts/install.sh

$ pip3 insntall -r requirements.txt
$ apt install sox libsox-fmt-all lame
$ mkdir record private audio

Get your api key here: https://platform.openai.com/account/api-keys

$ echo "{CHATGPT_ACCESS_KEY}" > private/api_keys

You can input text and send to ChatGPT through API
Then, you can hear the response

$ ./scripts/run_simple.sh

Start/Restart a ChatGPT session (wait for your voice audio file in the background)

$ ./scripts/start_background_session.sh

Stop the previous ChatGPT session if there is one

$ ./scripts/stop_background_session.sh

Start to record voice after it runs, ctrl+c when finished

$ ./scripts/record_audio.sh

Under Construction ...

• keyboard shortcut to record the user's voice
• keyboard shortcut to restart the ChatGPT session
• be able to load previous session from history
• ...

• OpenAI ChatGPT API Keys
  • https://platform.openai.com/account/api-keys
• OpenAI ChatGPT Python Chat Completions
  • https://platform.openai.com/docs/guides/chat
• Google Translate
  • https://py-googletrans.readthedocs.io/en/latest/
• OpenAI Whisper
  • https://github.com/openai/whisper