Recent advances in ASR Applied to an Arabic transcription system for Al-Jazeera

Patrick Cardinal; Ahmed Ali; Najim Dehak; Yu Zhang; Tuka Al Hanai; Yifan Zhang; James Glass; Stephan Vogel

Recent advances in ASR Applied to an Arabic transcription system for Al-Jazeera

Patrick Cardinal^*
, Ahmed Ali
, Najim Dehak
, Yu Zhang
, Tuka Al Hanai
, Yifan Zhang
, James Glass
, Stephan Vogel

^*Corresponding author for this work

Qatar Computing Research Institute

Research output: Contribution to journal › Conference article › peer-review

26 Citations (Scopus)

Abstract

This paper describes a detailed comparison of several state-of-the-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-vector-based 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%.

Original language	English
Pages (from-to)	2088-2092
Number of pages	5
Journal	Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH
Publication status	Published - 2014
Event	15th Annual Conference of the International Speech Communication Association: Celebrating the Diversity of Spoken Languages, INTERSPEECH 2014 - Singapore, Singapore Duration: 14 Sept 2014 → 18 Sept 2014

Keywords

ASR system
Arabic
Kaldi

Cite this

@article{5e6a56f622274140a1abb6836182be44,

title = "Recent advances in ASR Applied to an Arabic transcription system for Al-Jazeera",

abstract = "This paper describes a detailed comparison of several state-of-the-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-vector-based 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\%.",

keywords = "ASR system, Arabic, Kaldi",

author = "Patrick Cardinal and Ahmed Ali and Najim Dehak and Yu Zhang and \{Al Hanai\}, Tuka and Yifan Zhang and James Glass and Stephan Vogel",

note = "Publisher Copyright: Copyright {\textcopyright} 2014 ISCA.; 15th Annual Conference of the International Speech Communication Association: Celebrating the Diversity of Spoken Languages, INTERSPEECH 2014 ; Conference date: 14-09-2014 Through 18-09-2014",

year = "2014",

language = "English",

pages = "2088--2092",

journal = "Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH",

issn = "2308-457X",

publisher = "International Speech Communication Association",

}

TY - JOUR

T1 - Recent advances in ASR Applied to an Arabic transcription system for Al-Jazeera

AU - Cardinal, Patrick

AU - Ali, Ahmed

AU - Dehak, Najim

AU - Zhang, Yu

AU - Al Hanai, Tuka

AU - Zhang, Yifan

AU - Glass, James

AU - Vogel, Stephan

PY - 2014

Y1 - 2014

N2 - This paper describes a detailed comparison of several state-of-the-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-vector-based 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%.

AB - This paper describes a detailed comparison of several state-of-the-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-vector-based 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%.

KW - ASR system

KW - Arabic

KW - Kaldi

UR - https://www.scopus.com/pages/publications/84910093677

M3 - Conference article

AN - SCOPUS:84910093677

SN - 2308-457X

SP - 2088

EP - 2092

JO - Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH

JF - Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH

T2 - 15th Annual Conference of the International Speech Communication Association: Celebrating the Diversity of Spoken Languages, INTERSPEECH 2014

Y2 - 14 September 2014 through 18 September 2014

ER -

Recent advances in ASR Applied to an Arabic transcription system for Al-Jazeera

Abstract

Keywords

Other files and links

Fingerprint

Cite this