Voice Assistant For Personal Computers

Today almost every Operating System has some sort of voice assistant. For example, Siri, google(the google assistant), and Bixby. Even the windows system has one called Cortona. But we wanted to build a robust voice assistant for windows and Linux which could perform all the rudimentary tasks in the OS.

Required dependencies:

 pip install python3 
 pip install time
 pip install beautyfulsoup4
 pip install google
 pip install speechrecognition
 pip install pipwin
 pip install pyaudio
 pip install transformers
 pip install tensorflow

To Execute:

Run the following command: python3 project_voice_recognition.py
This will execute the file (Please install any other dependencies if prompted by the compiler)
The code will print “talk” then please give the command.
The code executes inside an infinite while loop, to stop it say “stop” after the “talk” command and the code will stop.

Methodology:

• We first created a voice-to-text algorithm on python using the library Speech_Recognition.
• For accessing the microphone we used the module speech_recognition. Microphone and to convert the speech to text, we use google_recognize(), it uses google's free web search API and does not require any API key.
• The output of the algorithm is a string of the user's speech.
• For text segmentation, we required a model which can produce good results without any prior training. This is because we do not have a dataset for voice commands specifically for Linux Voice assistants. So the solution to the problem was Zero-Shot Classification(ZSL). ZSL is a model which can learn from labeled data and predict the classes which were never seen during the training of the model.
• This model is most useful in our case as we just need to define certain classes of output and the model will label the input with the given classes.
• We used the ZSL model from an online repository- “Hugging face”. Hugging Face is an online community that provides models based on open-source code technologies. The model used for this project is: facebook/bart-large-mnli.
• Lastly we mapped all the labels of ZSL outputs to the Linux functions using python and the OS library of python.
• The OS library is used to implement functionalities like showing the contents of the folder and make a new directory.

Outcomes:

The voice assistant can run the following commands:

• Show contents of a file.
• Show today’s date.
• Search for a given keyword on google.
• Make a new directory.
• Simple arithmetic operations.

Team Members:

• Aasrish Vinay Perumalla (B20CS042)
• Hriday Kondru (B20CS021)
• Rohit Bhanudas Kote (B20CS056)
• Ribhav Manish (B20CS054)