Elamur Research⌕
☰
home / papers / 2007 / Introduction to Arabic Speech Recognition Using CMUSphinx System

Introduction to Arabic Speech Recognition Using CMUSphinx System

Paper · 2007 · H. Satori, M. Harti and N. Chenfour

Summary

The authors ask whether CMU Sphinx-4, an open-source HMM-based recognizer built mainly around English, can be adapted to Arabic. Arabic has received little ASR research compared with other major languages. It is also hard for ASR: written text usually lacks diacritics, dialects vary widely, and the morphology is rich. The paper first describes Sphinx-4's modular design, which has three parts: a FrontEnd, a Linguist (acoustic model, dictionary and language model) and a Decoder. It also covers SphinxTrain, the tool used to train acoustic models, and outlines Arabic phonology. As a proof of concept, the authors built Hello_Arabic_Digit, a recognizer for the ten Arabic digits. They replaced the three language-dependent components and trained on a small in-house corpus of Moroccan male speakers. Three men then each spoke the digits, and the authors computed a mean recognition ratio for each tester. They call the results very satisfactory given the tiny training set, though the extracted text does not include the actual figures. The work matters mainly as an early, low-budget recipe for bringing an English-centred toolkit to a new language.

Key points

  • The system is built on Sphinx-4, which is written in Java and was developed jointly by Carnegie Mellon, Sun Microsystems Laboratories and MERL. Acoustic models are trained with SphinxTrain, first publicly released on 7 June 2001.

  • The authors adapt three language-dependent components:

    • an acoustic model over phonemes,
    • a pronunciation dictionary,
    • a language model or grammar.

    Every word in the language model must also appear in the dictionary.

  • The training corpus has 300 tokens: 6 male Moroccan speakers × 10 digits × 5 repetitions. Each recording was played back to confirm the whole digit was captured, and all tokens were used for training.

  • Standard Arabic is described as having 34 phonemes: 6 vowels (3 short, 3 long) and 28 consonants. The allowed syllable shapes are CV, CVC and CVCC, and a word has as many syllables as vowels.

  • For evaluation, three men each uttered the 10 digits, and a mean recognition ratio was reported per tester. The paper does not state whether the test speakers overlap with the training speakers.

Interesting ideas

  • I1. Sphinx-4 treats the front end, the knowledge base and the search as separately replaceable modules, so researchers can test one alternative implementation without touching the rest of the system.
  • I2. A hard cross-resource rule says every word the language model can produce must have a pronunciation entry, which catches gaps between resources before decoding starts.
  • I3. Before scaling up, the authors validate the whole adaptation pipeline on a tiny closed vocabulary (ten digits), which keeps failures easy to trace.
  • I4. Each recording was checked by playback at collection time, so incomplete samples never entered training.
  • I5. Language-specific structural rules, such as the allowed syllable shapes and the one-vowel-per-syllable rule, are written down as explicit knowledge rather than left implicit in statistics.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (tree): Keep the perceptron parser and gparse behind a single interface so either can be swapped in, and their effect on frames and answers compared, without changing downstream code.
  • I2 (world): Add a similar build-time check so that every verb used by frame rules has a VerbNet class and an ontology node, and report any orphans explicitly.
  • I3 (answer): Pilot new reader types on a very small closed domain, such as counting problems, so that every wrong answer can be traced through tree, frames and world.
  • I4 (learn): Validate each silver item as it is created, for example by checking witness-parser agreement and well-formedness, instead of filtering the whole set afterwards.
  • I5 (tokens): Encode English morphological and orthographic constraints as explicit validators for the tagger and lemmatizer, so that impossible analyses are rejected with a stated reason.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSignal Processing · 452 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 2,908 worksSpeech & spoken language · 480 works

Cite

Cite as H. Satori, M. Harti and N. Chenfour. 2007. Introduction to Arabic Speech Recognition Using CMUSphinx System. arXiv:0704.2083.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.0704.2083 · arXiv 0704.2083

@misc{satori2007introduction,
  title = {{Introduction to Arabic Speech Recognition Using CMUSphinx System}},
  author = {H. Satori, M. Harti and N. Chenfour},
  year = {2007},
  doi = {10.48550/arxiv.0704.2083},
  eprint = {0704.2083},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.0704.2083}
}

References

20 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. A.Yousfi, “Introduction de la vitesse d’élocution et de l’énergie dans un modèle de reconnaissance automatique de la parole, ” Thèse de Doctorat, Faculté des Sciences Oujda, 2002. find
  2. S. Young, “The HTK hidden Markov model toolkit: Design and philosophy,” Cambridge University Engineering Department, UK, Tech. Rep. CUED/FINFENG/TR152, Sept. 1994. find
  3. N. Deshmukh, A. Ganapathiraju, J. Hamaker, J. Picone, and M. Ordowski, “A public domain speech-to-text system,” in Proc. 6th European Conf. Speech Communication and Technology, vol. 5, Budapest, Hungary, Sept. 1999, pp. 2127–2130. find
  4. X. X. Li, Y. Zhao, X. Pi, L. H. Liang, and A. V. Nefian, “Audio-visual continuous speech recognition using a coupled hidden Markov model,” in Proc. 7th International Conf. Spoken Language Processing, Denver, CO, Sept. 2002, pp. 213–216. find
  5. K. F. Lee, H. W. Hon, and R. Reddy, “An overview of the SPHINX speech recognition system,” IEEE Transactions on Acoustics, Speech and Signal Processing, vol. 38, no. 1, pp. 35–45, Jan. 1990. ↗
  6. X. Huang, F. Alleva, H. W. Hon, M. Y. Hwang, and R. Rosenfeld, “The SPHINX-II speech recognition system: an overview,” Computer Speech and Language, vol. 7, no. 2, pp. 137–148, 1993. ↗
  7. M. K. Ravishankar, “Efficient algorithms for speech recognition,” PhD Thesis (CMU Technical Report CS-96-143), Carnegie Mellon University, Pittsburgh, PA, 1996. find
  8. X.D. Huang, Y. Ariki, M.A. Jack, “Hidden Markov models for speech recognition,” Edinburgh: Edinburgh University Press, C, 1990. find
  9. K. Kirchho, J. Bilmes, J. Henderson, R. Schwartz, M. Noamany, P. Schone, G. Ji, S. Das, M. Egan, F. He, D. Vergyri, D. Liu, and N. Duta. 2002. Novel approaches to Arabic speech recognition. find
  10. D. Vergyri, K. Kirchhoff, K. Duh, A. Stolcke, "Morphology-based language modeling for Arabic speech recognition", In INTERSPEECH-2004, 2245-2248, 2004. find
  11. D. Vergyri, K. Kirchhoff. “Auto-matic diacritization of Arabic for acoustic modelling in speech recognition”. In Ali Farghaly andKarine Megerdoomian, editors, COLING 2004 Computational Approaches to Arabic Script-based Languages, pp. 66–73, Geneva, Switzerland, 2004. find
  12. G. Z. Hong “Speech Recognition Techniques for Digital Video Library,” University of Hong Kong , 2002. find
  13. X.D. Huang, “The SPHINX-II Speech Recognition System: An Overview,” Computer Speech and Language, Vol. 2, 1993; K. F. Lee, “Automatic Speech Recognition the Development of the SPHINX System,” Kluwer Academic Publishers, 1989. find
  14. M. Al-Zabibi, “An Acoustic–Phonetic Approach in Automatic Arabic Speech Recognition,” The British Library in Association with UMI, 1990. ↗
  15. A. Muhammad, “Alaswaat Alaghawaiyah,” Daar Alfalah, Jordan, 1990 (in Arabic). find
  16. J. Deller, J. Proakis, J.H. Hansen, “Discrete-Time Processing of Speech Signal,” Macmillan, NY, 1993. find
  17. M. Elshafei, “Toward an arabic text-to-speech system,” The Arabian J. Science and Engineering vol. 4B no. 16, pp. 565–583, 1991. find
  18. Y.A. El-Imam, “An unrestricted vocabulary arabic speech synthesis system”, IEEE Transactions on Acoustic, Speech, and Signal Processing vol. 37 , no. 12, pp.1829–1845, 1989. ↗
  19. G. Pullum, W. Ladusaw, Phonetic Symbol Guide, The University of Chicago Press, 1996. find
  20. Y. Ajami Alotaibi “Investigating spoken Arabic digits in speech recognition setting” Inf. and Comp. Sc. pp.173, 115, 2005. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/satori-2007-introduction-arabic-speech-recognition-using/