Everyone who speaks a language, speaks it with an accent. A particular accent essentially reflects a person's linguistic background. When people listen to someone speak with a different accent from their own, they notice the difference, and they may even make certain biased social judgments about the speaker.
Two networks have been implemented - CNN and LSTM.
We have trained only three languages: english, mandarin, arabic
Packages you need
Keras
Librosa
pydub
sklearn
numpy
pandas
Steps to run the project
First we need to download the required audio files from the speech accent archive database which is located here - accent.gmu.edu/browse_language.php
Create a folder inside data folder named as audio.
Go to code folder. And then run the getaudio.py file.
cd code
python getaudio.py
Wait till all the required audio is downloaded
Now, lets start training and predicting. If you want to run train with CNN, enter the following :
python main.py cnn 100
Where 100 is the number of epochs.
And, if you want to train with LSTM, enter the following:
python main.py lstm 100
NOTE: It is necessary to specify the network you want to train. However, if epochs are not specified, default will be 10.