Error analysis based on acoustic model has been performed in continuous Chinese speech recognition system Easy talk. Speaker adaptive training (SAT) and discriminative training with minimum phone frame error (MPFE) criterion has been used to decrease the errors in Finish Morph based continuous recognition system. Noise compensation technique has been implemented that results in 1.1% reduction of WER. Performance of English ASR has been analyzed based on word recognition error rate on subset of Malach corpus. Data of fifty two male and female speakers, 8440 utterance and connected digits has been recorded at different places having different SNR’s. Performance of English speech recognition has been evaluated in noisy condition by using HTK toolkit. Brief survey of ASR in Urdu and different languages will be presented in this section. Literature Review Many ASR systems has been developed using different methods on different languages. The next section describes the previously work done on Urdu ASR and of other languages. Speech corpus has been updated based on the results. Based on the test results confusion matrix has been generated indicating the correctly matched and confused phonemes. It consists of mixture of read and spontaneous speech and divided into two portions for training and testing. The speech corpus has been recorded based on earlier corpus design. Introduction This paper describes the implementation of single speaker ASR system and process employed in the analysis of speech corpus based on recognition results. The paper explores the possibility of analyzing the Urdu speech corpus based on recognition results to improve word error rate (WER). One of the major hurdles in such ASR is unacceptable word error rate. Sarmad Hussain Center for Language Engineering, Al-Khawarizmi Institute of Computer Science, University of Engineering and Technology, Lahore, Pakistan Abstract Speaker independent, spontaneous and continuous speech recognition system (ASR) can be integrated to other technologies like mobile to create an interface between technology and illiterate people so that they can use modern technologies. Error Analysis of Single Speaker Urdu Speech Recognition System Saad Irtza, Dr.

Urdu speech to text software