An NLP based approach to detect and identify languages word-by-word for audio data on local Pakistani languages
A spectogram-based approach to identifying langauges most commonly spoken in Pakistan. This includes Urdu, English, Arabic, Pashto, and Sindhi. The model is contructed as CNN with LSTM layers achieving an accuracy of 95%.
import the file languageLocalization.py to use in your script
def languageLocalize(inputFile, extension, chunk_file):
Parameters:
inputFile -- path to file for audio
extension -- audio file extension
chunk_file -- path to folder where the model stores preprocessing data
Return Value:
pred -- python array of strings containing the languauges predicted at positional interval
pred[i] is the language detected between (i - 1) and (i)th second
Example:
pred[3] = 'english' means english detected between 2-3 sec of audio
the notebook contains the script to run live detection on audio.
Tensorflow Keras Numpy Scipy Pydub Librosa