Pakistani Software Engineer Makes Worlds 1st Urdu Speech Recognition Application

The likelihood of some product application developer making an Urdu speech recognition program just got more probable as the most central instrument required for it has recently been created at Lahore’s Information Technology University.

Semantic innovation master Dr. Agha Ali Raza and his group at ITU’s Center for Speech and Language Technologies (CSaLT) research center has discharged for open utilize a corpus of Urdu sentences that covers all conceivable particular sounds, called phoneme by etymologists, utilized as a part of ordinary speech. This corpus including 708 sentences that covers each of the 63 phonemes will soon be accessible for download at the C-SALT site.

Those inspired by building up an Urdu speech recognition programming will now have entry to the most essential fixing required for the reason. They will simply require a store of words utilized as a part of ordinary speech to continue with building up the application.

Speech recognition is a two-stage handle. The corpus will give the PC application access to every conceivable phoneme utilized as a part of arrangement of significant Urdu words from regular speech. In spite of the fact that there are 63 particular phonemes in Urdu, in ordinary speech these don’t compare to 63 unmistakable sounds. Dr. Raza clarifies that sound made for a phoneme may shift starting with one articulation then onto the next relying upon the phoneme utilized previously, then after the fact it in a word. In this way, for each phoneme x, there will be 63*x*63 conceivable (tri-phoneme) sounds. The corpus of sentences spreads for all these conceivable sounds.

In the initial step, words from the corpus will permit the application to prepare itself in the hints of different Urdu words. The different vault of words will become an integral factor in the second stage permitting the application to pick the most proper words for the yield sentences.


In this manner, the precision of the speech recognition programming relies on upon composed or oral sources from where words and sentences are created for the corpus and the vault kept up independently to rule out aimless words.

Dr. Raza and his group has depended upon composed material got from daily paper and magazine articles in Urdu to create this corpus. In another corpus a work in progress at the CSaLT lab, he is utilizing oral material recorded with authorization from a few thousand telephonic discussions.


Please enter your comment!
Please enter your name here