Punjabi Speech Recognition
We are building the first Speech Recognition System for the Pakistani Punjabi Language which is also known as the SHAHMUKHI script. Unfortunately, there was almost zero work done before on Pakistani Punjabi so we had to start from scratch. We gathered 50k+ words of Pakistani Punjabi for the dictionary of our model. We extracted these words from different resources including Indian and Pakistani.
We have 10 speakers in our data collection part of which 3 are females and 7 are males. We deployed an Asterisk VoIP software also known as PBX for the recording of audios and for the annotation, we used different Indian news channel's articles. We convert the Indian Punjabi into Pakistani Punjabi using a converter and then we remove the grammatical mistakes manually. We are feeding our data to the Kaldi toolkit which uses 7 models of machine learning including sgmm2 and NNET2. We used G2P for making lexicons. We manually made lexicons of 1200+ words and feed the G2P as training data.
We are also going to build a mobile application in which we will deploy our model so that a novice can use it without any difficulty. For this, we are trying to use a wrapper or to use XLSR which is one of the latest technology by Facebook, if needed.
Keywords: Machine Learning,Development,Artificial Intelligence,Mobile Application,Android,Web Application
Tools: Python,Asterisk,Kaldi,Android Studio,React Native,VoIP,Server,XLSR
Department: Department of Computer Science
Project Team Members
| Name |
Email |
|
Syed Uzair Ul Hassan
|
suzair2018@namal.edu.pk |
|
Zeemal Urooj
|
zeemal2018@namal.edu.pk |
|
Muhammad Sajid Hameed
|
msajid2018@namal.edu.pk |
Project Poster
