Recent advances in ASR applied to an Arabic transcription system for Al-Jazeera
نویسندگان
چکیده
This paper describes a detailed comparison of several state-ofthe-art speech recognition techniques applied to a limited Arabic broadcast news dataset. The different approaches were all trained on 50 hours of transcribed audio from the Al-Jazeera news channel. The best results were obtained using i-vectorbased speaker adaptation in a training scenario using the Minimum Phone Error (MPE) criteria combined with sequential Deep Neural Network (DNN) training. We report results for two different types of test data: broadcast news reports, with a best word error rate (WER) of 17.86%, and a broadcast conversations with a best WER of 29.85%. The overall WER on this test set is 25.6%.
منابع مشابه
Critique of Arabic Texts/ Critical Review of Al-Adab al-Arabi Min-Asr al-Jahili, Ali Ghahramani and Masoud Bavanpour
متن کامل
Recent Advances in T Cell Signaling in Aging
The immune system of mammalian organisms undergoes alterations that may account for an increased susceptibility to certain infections, autoimmune diseases, or malignancies. Well characterized are age related defect in T cell functions and cell mediated immunity. Although it is well established that the functional properties of T cells decrease with age, its biochemical and molecular nature is...
متن کاملAutomated Speech Recognition System (ASR)
This paper reports the results of the first phase of a research work for building a high performance, speakerindependent natural Arabic speech recognition system. This work aims at developing an Arabic broadcast news transcription system and a base system for further research. Several concurrent recent advances in Arabic language processing were crucial for the success of this stage, e.g automa...
متن کاملRevisiting the Arabic Diglossic Situation and Highlighting the Socio-Cultural Factors Shaping Language Use in Light of Auer’s (2005) Model
In the field of Arabic sociolinguistics, diglossia has been an interesting linguistic inquiry since it was first discussed by Ferguson in 1959. Since then, diglossia has been discussed, expanded, and revisited by Badawi (1973), Hudson (2002), and Albirini (2016) among others. While the discussion of the Arabic diglossic situation highlights the existence of two separate codes (High and Lo...
متن کاملAdvances in the CMU/Interact Arabic GALE Transcription System
* Now with Toshiba Research Europe Ltd, Cambridge, United Kingdom ABSTRACT This paper describes the CMU/InterACT effort in developing an Arabic Automatic Speech Recognition (ASR) system for broadcast news and conversations within the GALE 2006 evaluation. Through the span of 9 month in preparation for this evaluation we improved our system by 40% relative compared to our legacy system. These im...
متن کامل