VoiceGPT

• This is a basic python program which recognizes user's voice and computes it with the AI to get the response. The response is converted to speech and then played back to the user!
• It has 2 different models, one is based on EdgeGPT and the other is based on LLaMa.cpp
• EdgeGPT uses browser cookies and gets the response from the cloud where Bing's new chat feature is released. LLaMa on the other hand gets the response from the local API running on the user's computer.
• The speed of everything here depends upon your PC specs. As the Voice recognition uses OpenAI's Tiny Whisper Model that will be running on your PC to get the voice data.
• Both GPUs and CPUs are supported, However GPUs are preferred more due to their speed. CPUs can work too but they will be kind of slower compared to GPU computing.
• The Project is fully customisable! as it has a config file which has all the settings necessary to run the project on your local computer.

Requirements

• You need Python 3.9 as Pytorch doesn't support the latest versions.
• You need a good PC to run the models. Mid range hardware with minimal processor will also work.
• You should have access to Bing Chat.
• Microsoft Bing browser to get the cookies.

Installation

First you need to clone the repository using git clone.
git clone https://github.com/Shreyas-ITB/VoiceGPT
Or just download the latest release from the releases page

Now you need cookies from your Microsoft Edge browser. To get that open your Edge browser and install this extension.

• for chrome or chromium based browsers
• for firefox
• Go to bing.com on your microsoft browser
• Open the extension
• Click "Export" on the bottom right, then "Export as JSON" (This saves your cookies to clipboard)
• Go to the config directory, open the cookies.json file and paste your cookie data.

In the terminal or CMD enter the below command to install the project module requirements.
pip install -r ./config/requirements.txt
or on windows
pip install -r .\config\requirements.txt

Now you need to configure the config.ini file as per your prefrence.

• The config.ini file is located in the config directory.

[Default]
# Leave this default if your model file name is the same.
# Enter your model file name in the below field.

ModelFilename = ggml-model-q4_1.bin

# Leave this default if you don't want a different port.
# Enter a different port number in which the API will be hosted.

APIPort = 8080

# Enter the API Response call wait time in seconds here.
# This let's the API know when to get back the response, This also depends on your system RAM.
# If you have more RAM then probably it will be 5 seconds, less RAM then higher than 5 seconds.

ResponseWaitTime = 5

# If you want to run LLaMa on your local machine then specify the below field as yes.
# If you only want EdgeGPT then you can leave it as it is or just specify the below field as no.

LLaMainit = no

# If you don't want Voice input to the AI (controlled by voice) then you can leave it as it is.
# However if you want the Voice input then change the below setting to voice instead of text.
# Voice will activate the AI using your voice. Text will allow you to type the prompt!

Inputmode = text

If you want to use Alpaca, Vicunia or LLaMa model locally then you can download the model from huggingface and paste it in the models directory.
If you have a .pth file then you can use the convert-pth-to-ggml.py file to convert it. Note that only ggml format models can run.
Also note that the ModelFilename should be the exact file name of your model that you added in the models directory.

Now its time to compile the C++ files.

• On linux or mac, open up a terminal in the project's directory, and enter this command.
• make
• On Windows you might have to install cmake
• After that just open up CMD in the project's directory and enter the below command.
• cmake .

Once you have followed the previous steps, you can now proceed to run the project.

• To do so, in the terminal just enter the below command.
• python3 main.py
• Or if you have windows just enter this command.
• python main.py

If you downloaded the project from the releases page just run the executable.

• On linux and mac
• ./main
• On windows
• main.exe

And there you have it. Now you can talk to your very own AI!

Projects used

• OpenAI Whisper
• LLaMa.cpp
• EdgeGPT
• Alpaca.http
• Alpaca.cpp

Contributions

Any contributions, bug reports or pull requests with good intention will be always appreciated.