Study on pharyngeal and uvular consonants in foreign accented Arabic for ASR

https://doi.org/10.1016/j.csl.2009.04.005Get rights and content

Abstract

This paper investigates the unique pharyngeal and uvular consonants of Arabic from the point of view of automatic speech recognition (ASR). Comparisons of the recognition error rates for these phonemes are analyzed in five experiments that involve different combinations of native and non-native Arabic speakers. The most three confusing consonants for every investigated consonant are discussed. All experiments use the Hidden Markov Model Toolkit (HTK) and the Language Data Consortium (LDC) WestPoint Modern Standard Arabic (MSA) database. Results confirm that these Arabic distinct consonants are a major source of difficulty for Arabic ASR. While the recognition rate for certain of these unique consonants such as /ℏ/ can drop below 35% when uttered by non-native speakers, there is advantage to include non-native speakers in ASR. Besides, regional differences in pronunciation of MSA by native Arabic speakers require the attention of Arabic ASR research.

Section snippets

Introduction and background

Arabic is a Semitic language which has many differences when compared with Indo-European languages such as English. Some of the differences include unique phonemes and phonetic features, and a complicated morphological word structure. It has been shown that major difficulties in automatic speech recognition (ASR) systems dedicated to Modern Standard Arabic (MSA) can be attributed to distinctive characteristics of the Arabic sound system, namely, geminate, emphatic, uvular, and pharyngeal

Experimental framework

The system presented in this paper is designed to recognize Arabic phonemes. In this investigation we analyze the performance of the system with respect to the pharyngeal consonants – /ʕ/ and /ℏ/ – and the uvular consonants /ɣ/, /q/, and /x/. The study focuses on the effect of native and non-native speakers in both training and testing data. The accuracies with respect to all five consonants and in all conducted experiments are reported and investigated in detail. The effect of the mother

Results

The results reported here are based on the outcomes of the Arabic ASR system described above. This system computed the accuracies of all Arabic phonemes without using any LM. Five experiments were carried out in this investigation. These experiments differ only in the type of the training and testing data sets. These experiments are labeled as N/N, N/NN, NN/N, NN/NN, and M/M. In the experiments, N/N indicates that native Arabic speakers are used in both training and testing phases. Native

Conclusion

An Arabic phoneme recognition system was designed and used to investigate Arabic pharyngeal and uvular phonemes in Modern Standard Arabic (MSA). The investigation depended mainly on speech recognition outcomes. This speech recognition recognizes speech signal by using phoneme level without using any language model. The most three confusing phonemes that degraded accuracy of every phoneme in our set were presented and discussed. The most important outcome from all manipulation presented in this

References (34)

  • M.M. Awais et al.

    Arabic phoneme identification using conventional and concurrent neural networks in nonnative speakers

    Lecture Notes in Computer Science (LNCS)

    (2007)
  • K.M. Bisher

    Arabic Phonetics

    (1990)
  • Borden, G.J., Harris, K.S., 1990. Speech Science Primer. Dar Al-Sharq Al-Arabi (translated and expanded to Arabic...
  • P. Delattre

    Pharyngeal features in the consonants of Arabic, German, Spanish, French and American English

    Phonetica

    (1971)
  • J. Deller et al.

    Discrete-time Processing of Speech Signal

    (1993)
  • Elgendy, A.M., 2001. Aspects of Pharyngeal Coarticualtion. Ph.D. Dissertation, University of Amsterdam, LOT 44,...
  • Y. El-Imam

    An unrestricted vocabulary Arabic speech synthesis system

    IEEE Transactions on Acoustic, Speech, and Signal Processing

    (1989)

    • Cited by (13)

    • Speaker recognition based on Arabic phonemes

      2017, Speech Communication
      Citation Excerpt :

      In Arabic, a vowel always forms the nucleus of a syllable; there are as many syllables within a word as there are vowels (Alghamdi, 2015). The Arabic alphabet consists of 28 consonants and six vowels (three long and three short vowels) (Alghamdi, 2015; Alotaibi and Muhammad, 2010; Versteegh, 2014; Jalt, 2016). The three short vowels are represented by the phonemes /i/, /a/, and /u/, and the three long vowels are represented by the phonemes /i:/, /a:/, and /u:/ (Versteegh, 2014).

    • Arabic speaker recognition system based on phoneme fusion

      2024, Multimedia Tools and Applications

    • Modern Standard Arabic Speech Corpora: A Systematic Review

      2023, IEEE Access

    • Spoken Language Identification System Using Convolutional Recurrent Neural Network

      2022, Applied Sciences (Switzerland)

    • Phonological and lexical abilities of Jordanian children with autism

      2020, Applied Neuropsychology: Child

    • Unsupervised Phonetic Segmentation of Classical Arabic Speech Using Forward and Inverse Characteristics of the Vocal Tract

      2020, Arabian Journal for Science and Engineering
    View all citing articles on Scopus
    View full text
    Copyright © 2009 Elsevier Ltd. All rights reserved.