---
title: "Syllable-Based Speech Recognition for Amharic — Abate 2007"
description: "The paper builds an automatic speech recognizer for Amharic, a Semitic language with few speech tools. 25 references, cited by 1 paper here."
authors: ["Solomon Teferra Abate", "Wolfgang Menzel"]
year: 2007
venue: "Proceedings of the 2007 Workshop on Computational Approaches to Semitic Languages: Common Issues and Resources"
source: "https://aclanthology.org/W07-0805/"
licence: "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)"
licence_url: "https://creativecommons.org/licenses/by-nc-sa/3.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/abate-2007-syllable-based-speech-recognition-amharic/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/abate-2007-syllable-based-speech-recognition-amharic/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2007](https://md.research.elamur.ai/free/en/papers/2007/page/2/) / Syllable-Based Speech Recognition for Amharic

# Syllable-Based Speech Recognition for Amharic

**Authors:** [Solomon Teferra Abate](https://md.research.elamur.ai/free/en/authors/solomon-teferra-abate/) and [Wolfgang Menzel](https://md.research.elamur.ai/free/en/authors/wolfgang-menzel/) · **Year:** 2007 · **Venue:** [Proceedings of the 2007 Workshop on Computational Approaches to Semitic Languages: Common Issues and Resources](https://md.research.elamur.ai/free/en/venues/workshop-on-computational-approaches-to-semitic-languages-common-issues-and/) ([Workshop on Computational Approaches to Semitic Languages: Common Issues and Resources 2007](https://md.research.elamur.ai/free/en/venues/workshop-on-computational-approaches-to-semitic-languages-common-issues-and/2007/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY-NC-SA 3.0 (ACL Anthology, before 2016)](https://creativecommons.org/licenses/by-nc-sa/3.0/)

**Links:** [Publisher's page](https://aclanthology.org/W07-0805/) · [PDF](https://aclanthology.org/W07-0805.pdf) · [OpenAlex](https://openalex.org/W2012026885)

**Citations:** 24 citations (OpenAlex, 2026-10-14) · [1](https://md.research.elamur.ai/free/en/papers/abate-2007-syllable-based-speech-recognition-amharic/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper builds an automatic speech recognizer for Amharic, a Semitic language with few speech tools. Amharic script maps almost one-to-one onto consonant-vowel sounds. The authors use this property to choose CV syllables, not phones, as the acoustic units. After merging graphemes that sound the same, the 276-symbol fidel reduces to 233 distinct CV syllables, and each one gets its own HMM. The system also uses bigram language models trained with HTK and an existing syllable-level pronunciation dictionary. The authors vary the number of emitting states, the skip/jump transitions and the Gaussian mixtures per state. The best setup has five emitting states, no skips and 12 mixtures. With speaker adaptation it reaches 90.43% word accuracy on a 5k evaluation set. A word-internal triphone baseline scores somewhat higher (91.31%), but the syllable system is much smaller and faster. This shows that syllable units are a practical choice for languages whose script is syllabic.

## Key points
- The Amharic writing system has 276 symbols. Removing redundant graphemes leaves 233 distinct CV syllables, and each syllable is modelled by a separate HMM.
- The pronunciation dictionaries come from Solomon et al. (2005): a canonical one covering 50,000 words and an alternative one covering 25,000 words, both transcribed in CV syllables.
- These dictionaries do not mark three things:
  - gemination;
  - whether the sixth-order vowel is actually pronounced;
  - whether the glottal stop is present.
- The best topology on the 5k development set is 5 emitting states, 12 Gaussians and no skips, scoring 89.80% (87.69% on 20k). On the 5k evaluation set it reaches 90.43% with LM and speaker adaptation.
- Compared with the triphone baseline (91.31%), the syllable recognizer needs 15MB instead of 38MB and runs 37% faster. Unlike the baseline, it uses no state tying.

## Interesting ideas
- Choose the modelling unit that the writing system already provides, and merge graphemes that are written differently but sound the same before modelling. This gives a compact, well-sized unit inventory.
- Give each unit model capacity in proportion to how often it occurs: frequent syllables get up to 58 Gaussians, rare ones at most 2. This avoids both underfitting common units and overfitting rare ones.
- One transcription symbol covering two different realizations forces a single model to learn two sounds. The authors trace the remaining errors to this representation problem rather than to the model.
- Model selection weighs memory, number of Gaussians and decoding speed alongside accuracy, and accepts a slightly less accurate system that is much cheaper.
- Topologies are tuned on development sets, and only the top two are rechecked on separate evaluation sets to confirm that the ranking holds.

## For Mova

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

- **I1** (tokens): Collapse orthographic variants (British/American spellings, typographic apostrophes, ligatures) into one canonical unit before tagging and lemmatization, so rare surface forms share statistics.
- **I2** (learn): When training the perceptron or gparse, scale feature budgets or per-class parameters by frequency (for example per lemma or per dependency relation), so rare labels are not overparameterized on thin silver data.
- **I3** (tree): When a UD label or POS tag merges two behaviours that act differently, split it internally before training and map it back to the standard label at output; error analysis should check for such merged categories.
- **I4** (learn): Evaluation reports for the perceptron parser versus gparse should list size and speed next to LAS, so a cheaper explainable model can be chosen when the accuracy gap is small.
- **I5** (learn): Freeze candidates on dev and confirm only a short list on held-out test data, so self-induced repair rules are not tuned to the test set.

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

## Institutions

- [Universität Hamburg](https://md.research.elamur.ai/free/en/universities/universitat-hamburg/) — University, Hamburg, Germany · 81 works · 5,549 citations

📚 Shelves (7): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works · [💭 Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/) · 1,649 works · [💭 Experimental and Cognitive Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/experimental-and-cognitive-psychology/) · 930 works · [🔊 Signal Processing](https://md.research.elamur.ai/free/en/shelves/computer-science/signal-processing/) · 493 works

🧭 Topics (2): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 3,066 works · [🎙️ Speech & spoken language](https://md.research.elamur.ai/free/en/topics/speech/) · 519 works

## Cite

Cite as: Solomon Teferra Abate and Wolfgang Menzel. 2007. Syllable-Based Speech Recognition for Amharic. Proceedings of the 2007 Workshop on Computational Approaches to Semitic Languages: Common Issues and Resources.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{abate2007syllable,
  title = {{Syllable-Based Speech Recognition for Amharic}},
  author = {Solomon Teferra Abate and Wolfgang Menzel},
  booktitle = {Proceedings of the 2007 Workshop on Computational Approaches to Semitic Languages: Common Issues and Resources},
  year = {2007},
  doi = {10.3115/1654576.1654583},
  url = {https://doi.org/10.3115/1654576.1654583}
}
```

- [← previous SemEval-2007 Task 18: Arabic Semantic Labeling](https://md.research.elamur.ai/free/en/papers/diab-2007-semeval-2007-task-18-arabic/)
- [next → LCC-WSD: System Description for English Coarse Grained All Words Task at SemEval 2007](https://md.research.elamur.ai/free/en/papers/novischi-2007-lcc-wsd-system-description-english/)
- [↑ back to papers of 2007](https://md.research.elamur.ai/free/en/papers/2007/page/2/)

## References

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

1. Afif, Mohamed, Long Nguyen, Bing Xiang, Sherif Abdou, and John Makhoul. 2005. Recent progress in Arabic broadcast news transcription at BBN. In INTER SPEECH2005, 1637-1640 [find](https://scholar.google.com/scholar?q=Afif%2C+Mohamed%2C+Long+Nguyen%2C+Bing+Xiang%2C+Sherif+Abdou%2C+and+John+Makhoul%2E+2005%2E+Recent+progress+in+Arabic+broadcast+news+transcription+at+BBN%2E+In+INTER+SPEECH2005%2C+1637%2D1640)
2. Baye Yimam and TEAM 503 students. 1997. "ፊደል እንደገና" Ethiopian Journal of Languages and Literature 7(1997): 1-32. [find](https://scholar.google.com/scholar?q=Baye+Yimam+and+TEAM+503+students%2E+1997%2E+%22%E1%8D%8A%E1%8B%B0%E1%88%8D+%E1%8A%A5%E1%8A%95%E1%8B%B0%E1%8C%88%E1%8A%93%22+Ethiopian+Journal+of+Languages+and+Literature+7%281997%29%3A+1%2D32%2E)
3. Baye Yimam. 1986. "የአማርኛ ሰዋሰው". Addis Ababa. ት.መ.ማ.ማ.ድ. [find](https://scholar.google.com/scholar?q=Baye+Yimam%2E+1986%2E+%22%E1%8B%A8%E1%8A%A0%E1%88%9B%E1%88%AD%E1%8A%9B+%E1%88%B0%E1%8B%8B%E1%88%B0%E1%8B%8D%22%2E+Addis+Ababa%2E+%E1%89%B5%2E%E1%88%98%2E%E1%88%9B%2E%E1%88%9B%2E%E1%8B%B5%2E)
4. Bender, L.M. and Ferguson C. 1976. The Ethiopian Writing System. In Language in Ethiopia. Edited by M.L. Bender, J.D. Bowen, R.L. Cooper, and C.A. Ferguson. London: Oxford University press. [find](https://scholar.google.com/scholar?q=Bender%2C+L%2EM%2E+and+Ferguson+C%2E+1976%2E+The+Ethiopian+Writing+System%2E+In+Language+in+Ethiopia%2E+Edited+by+M%2EL%2E+Bender%2C+J%2ED%2E+Bowen%2C+R%2EL%2E+Cooper%2C+and+C%2EA%2E+Ferguson%2E+London%3A+Oxford+University+press%2E)
5. Cowley, Roger, Marvin L. Bender and Charles A. Fer-gusone. 1976. The Amharic Language-Description. In Language in Ethiopia. Edited by M.L. Bender, J.D. Bowen, R.L. Cooper, and C.A. Ferguson. London: Oxford University press. Deller, J.R. Jr., Hansen, J.H.L. and Proakis, J.G., Discrete-time Processing of Speech Signals. [find](https://scholar.google.com/scholar?q=Cowley%2C+Roger%2C+Marvin+L%2E+Bender+and+Charles+A%2E+Fer%2Dgusone%2E+1976%2E+The+Amharic+Language%2DDescription%2E+In+Language+in+Ethiopia%2E+Edited+by+M%2EL%2E+Bender%2C+J%2ED%2E+Bowen%2C+R%2EL%2E+Cooper%2C+and+C%2EA%2E+Ferguson%2E+London%3A+O)
6. Macmillan Publishing Company, New York, 2000. [find](https://scholar.google.com/scholar?q=Macmillan+Publishing+Company%2C+New+York%2C+2000%2E)
7. Fukada, Toshiaki, Takayoshi Yoshimura and Yoshinori Sagisa. 1999. Automatic generation of multiple pronunciations based on neural networks. Speech Communication 27:63—73 [http://citeseer.ist.psu.edu/fukada99automatic.html](http://citeseer.ist.psu.edu/fukada99automatic.html). [↗](http://citeseer.ist.psu.edu/fukada99automatic.html)
8. Ganapathiraju, Aravind; Jonathan Hamaker; Mark Ordowski; and George R. Doddington. 1997. Joseph Picone. Syllable-based Large Vocabulary Continuous Speech Recognition. [find](https://scholar.google.com/scholar?q=Ganapathiraju%2C+Aravind%3B+Jonathan+Hamaker%3B+Mark+Ordowski%3B+and+George+R%2E+Doddington%2E+1997%2E+Joseph+Picone%2E+Syllable%2Dbased+Large+Vocabulary+Continuous+Speech+Recognition%2E)
9. Getachew Haile. 1967. The Problems of the Amharic Writing System. A paper presented in advance for the interdisciplinary seminar of the Faculty of Arts and Education. HSIU. [find](https://scholar.google.com/scholar?q=Getachew+Haile%2E+1967%2E+The+Problems+of+the+Amharic+Writing+System%2E+A+paper+presented+in+advance+for+the+interdisciplinary+seminar+of+the+Faculty+of+Arts+and+Education%2E+HSIU%2E)
10. Hayward, Katrina and Richard J. Hayward. 1999. Amharic. In Handbook of the International Phonetic Association: A guide to the use of the International Phonetic Alphabet. Cambridge: the University Press. [find](https://scholar.google.com/scholar?q=Hayward%2C+Katrina+and+Richard+J%2E+Hayward%2E+1999%2E+Amharic%2E+In+Handbook+of+the+International+Phonetic+Association%3A+A+guide+to+the+use+of+the+International+Phonetic+Alphabet%2E+Cambridge%3A+the+University+Pres)
11. Hu, Zhihong; Johan Schalkwyk; Etienne Barnard; and Ronald Cole. 1996. Speech recognition using syllable like units. Proc. Int'l Conf. on Spoken Language Processing (ICSLP), 2:426-429. [find](https://scholar.google.com/scholar?q=Hu%2C+Zhihong%3B+Johan+Schalkwyk%3B+Etienne+Barnard%3B+and+Ronald+Cole%2E+1996%2E+Speech+recognition+using+syllable+like+units%2E+Proc%2E+Int%27l+Conf%2E+on+Spoken+Language+Processing+%28ICSLP%29%2C+2%3A426%2D429%2E)
12. Kanokphara, Supphanat; Virongrong Tesprasit and Rachod Thongprasirt. 2003. Pronunciation Variation Speech Recognition Without Dictionary Modification on Sparse Database, IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2003, Hong Kong). [find](https://scholar.google.com/scholar?q=Kanokphara%2C+Supphanat%3B+Virongrong+Tesprasit+and+Rachod+Thongprasirt%2E+2003%2E+Pronunciation+Variation+Speech+Recognition+Without+Dictionary+Modification+on+Sparse+Database%2C+IEEE+International+Conference+)
13. Kinfe Tadesse. 2002. Sub-word Based Amharic Word Recognition: An Experiment Using Hidden Markov Model (HMM), M.Sc Thesis. Addis Ababa University Faculty of Informatics. Addis Ababa. [find](https://scholar.google.com/scholar?q=Kinfe+Tadesse%2E+2002%2E+Sub%2Dword+Based+Amharic+Word+Recognition%3A+An+Experiment+Using+Hidden+Markov+Model+%28HMM%29%2C+M%2ESc+Thesis%2E+Addis+Ababa+University+Faculty+of+Informatics%2E+Addis+Ababa%2E)
14. Lee, C-H., Gauvain, J-L., Pieraccini, R. and Rabiner, L. R.. 1992. Large vocabulary speech recognition using subword units. Proc. ICSST-92, Brisbane, Australia, pp. 342-353. [find](https://scholar.google.com/scholar?q=Lee%2C+C%2DH%2E%2C+Gauvain%2C+J%2DL%2E%2C+Pieraccini%2C+R%2E+and+Rabiner%2C+L%2E+R%2E%2E+1992%2E+Large+vocabulary+speech+recognition+using+subword+units%2E+Proc%2E+ICSST%2D92%2C+Brisbane%2C+Australia%2C+pp%2E+342%2D353%2E)
15. Leslau, W. 2000. Introductory Grammar of Amharic, Wiesbaden: Harrassowitz. [find](https://scholar.google.com/scholar?q=Leslau%2C+W%2E+2000%2E+Introductory+Grammar+of+Amharic%2C+Wiesbaden%3A+Harrassowitz%2E)
16. Martha Yifiru. 2003. Application of Amharic speech recognition system to command and control computer: An experiment with Microsoft Word, M.Sc Thesis. Addis Ababa University Faculty of Informatics. Addis Ababa. [find](https://scholar.google.com/scholar?q=Martha+Yifiru%2E+2003%2E+Application+of+Amharic+speech+recognition+system+to+command+and+control+computer%3A+An+experiment+with+Microsoft+Word%2C+M%2ESc+Thesis%2E+Addis+Ababa+University+Faculty+of+Informatics%2E+Ad)
17. Rabiner, L. and Juang, B. 1993. Fundamentals of speech recognition. Englewood Cliffs, NJ. [find](https://scholar.google.com/scholar?q=Rabiner%2C+L%2E+and+Juang%2C+B%2E+1993%2E+Fundamentals+of+speech+recognition%2E+Englewood+Cliffs%2C+NJ%2E)
18. Hussien Seid and Björn. Gambäck 2005. A Speaker Independent Continuous Speech Recognizer for Amharic. In: INTERSPEECH 2005, 9th European Conference on Speech Communication and Technology. Lisbon, September 4-9. [find](https://scholar.google.com/scholar?q=Hussien+Seid+and+Bj%C3%B6rn%2E+Gamb%C3%A4ck+2005%2E+A+Speaker+Independent+Continuous+Speech+Recognizer+for+Amharic%2E+In%3A+INTERSPEECH+2005%2C+9th+European+Conference+on+Speech+Communication+and+Technology%2E+Lisbon%2C+Sept)
19. Solomon Birihanu. 2001. Isolated Amharic Consonant-Vowel (CV) Syllable Recognition, M.Sc Thesis. Addis Ababa University Faculty of Informatics. Addis Ababa. [find](https://scholar.google.com/scholar?q=Solomon+Birihanu%2E+2001%2E+Isolated+Amharic+Consonant%2DVowel+%28CV%29+Syllable+Recognition%2C+M%2ESc+Thesis%2E+Addis+Ababa+University+Faculty+of+Informatics%2E+Addis+Ababa%2E)
20. Solomon Teferra Abate. 2006. Automatic Speech Recognition for Amharic. Ph.D. Thesis. University of Hamburg. Hamburg. [find](https://scholar.google.com/scholar?q=Solomon+Teferra+Abate%2E+2006%2E+Automatic+Speech+Recognition+for+Amharic%2E+Ph%2ED%2E+Thesis%2E+University+of+Hamburg%2E+Hamburg%2E)
21. Solomon Teferra Abate, Wolfgang Menzel and Bairu Tafla. 2005. An Amharic Speech Corpus for Large Vocabulary Continuous Speech Recognition. In: IN-TERSPEECH 2005, 9th European Conference on Speech Communication and Technology. Lisbon, September 4-9. [find](https://scholar.google.com/scholar?q=Solomon+Teferra+Abate%2C+Wolfgang+Menzel+and+Bairu+Tafla%2E+2005%2E+An+Amharic+Speech+Corpus+for+Large+Vocabulary+Continuous+Speech+Recognition%2E+In%3A+IN%2DTERSPEECH+2005%2C+9th+European+Conference+on+Speech+Comm)
22. Tadesse Beyene. 1994. The Ethiopian Writing System. Paper presented at the 12 th International Conference of Ethiopian Studies, Michigan State University. [find](https://scholar.google.com/scholar?q=Tadesse+Beyene%2E+1994%2E+The+Ethiopian+Writing+System%2E+Paper+presented+at+the+12+th+International+Conference+of+Ethiopian+Studies%2C+Michigan+State+University%2E)
23. Wu, Su-Lin. 1998. Incorporating Information from Syllable-length Time Scales into Automatic Speech Recognition. PhD thesis, University of California, Berkeley, CA. [find](https://scholar.google.com/scholar?q=Wu%2C+Su%2DLin%2E+1998%2E+Incorporating+Information+from+Syllable%2Dlength+Time+Scales+into+Automatic+Speech+Recognition%2E+PhD+thesis%2C+University+of+California%2C+Berkeley%2C+CA%2E)
24. Young, Steve; Dan Kershaw; Julian Odell and Dave Ollason. 2002. The HTK Book. [find](https://scholar.google.com/scholar?q=Young%2C+Steve%3B+Dan+Kershaw%3B+Julian+Odell+and+Dave+Ollason%2E+2002%2E+The+HTK+Book%2E)
25. Zegaye Seyifu. 2003. Large vocabulary, speaker independent, continuous Amharic speech recognition, M.Sc Thesis. Addis Ababa University Faculty of Informatics. Addis Ababa. [find](https://scholar.google.com/scholar?q=Zegaye+Seyifu%2E+2003%2E+Large+vocabulary%2C+speaker+independent%2C+continuous+Amharic+speech+recognition%2C+M%2ESc+Thesis%2E+Addis+Ababa+University+Faculty+of+Informatics%2E+Addis+Ababa%2E)

*Updated 2 November 2026*