speech_mnist

Our speech mnist helps us in turning audio of numbers 0-9 into their digits. This app allows you to record a voice and say a number between ziro and nine and it will predict what your number is. It also will detect multiple numbers and seperate every number and gives you every spoken number.

pre-requesties

For using this example, you should install some packages. These packages are being used for generating our model

pip install numpy tensorflow pandas matplotlib scipy scikit-learn

To run this application, we use streamlit. Also for recording voices we use an open source repository named audio_recorder_streamlit. to wirk with wav file type we use pydub. To install these packages we will use the following command.

pip install pydub audio_recorder_streamlit streamlit

How to Use

After installing the pre-requesties, you should create the model. You could either train the model or download it from my drive. To train the model, you should download the dataset from Kaggle. Then put the data folder in a folder named Audio. Now you should run the data_loader.ipynb notebook line by line. Be careful that this notebook has high usage of RAM and CPU. After training our model, You could run our application with the following command

streamlit run audio_listner.py

A web page like this will be open on address http://localhost:8501/ on your browser With pressing on mic logo, your voice will be recorded. After that, our application will give you the numbers you have said. The result will be something like this