The Experts below are selected from a list of 360 Experts worldwide ranked by ideXlab platform

Million Meshesha - One of the best experts on this subject based on the ideXlab platform.

  • Amharic OCR : An End-to-End Learning
    Applied Sciences, 2020
    Co-Authors: Birhanu Belay, Marcus Liwicki, Tewodros Habtegebrial, Million Meshesha, Gebeyehu Belay, Didier Stricker
    Abstract:

    In this paper, we introduce an end-to-end Amharic text-line image recognition approach based on recurrent neural networks. Amharic is an indigenous Ethiopic script which follows a unique syllabic w ...

  • experimenting statistical machine translation for ethiopic semitic languages the case of Amharic tigrigna
    Information and Communication Technology for Development for Africa. First International Conference ICT4DA 2017 Bahir Dar Ethiopia September 25–27 201, 2018
    Co-Authors: Michael Melese Woldeyohannis, Million Meshesha
    Abstract:

    In this research an attempt have been made to experiment on Amharic-Tigrigna machine translation for promoting information sharing. Since there is no Amharic-Tigrigna parallel text corpus, we prepared a parallel text corpus for Amharic-Tigrigna machine translation system from religious domain specifically from bible. Consequently, the data preparation involves sentence alignment, sentence splitting, tokenization, normalization of Amharic-Tigrigna parallel corpora and then splitting the dataset into training, tuning and testing data. Then, Amharic-Tigrigna translation model have been constructed using training data and further tuned for better translation. Finally, given target language model, the Amharic-Tigrigna translation system generates a target output with reference to translation model using word and morpheme as a unit. The result we found from the experiment is promising to design Amharic-Tigrigna machine translation system between resource deficient languages. We are now working on post-editing to enhance the performance of the bi-lingual Amharic-Tigrigna translator.

  • a corpus for Amharic english speech translation the case of tourism domain
    International Conference on Information and Communication Technology, 2017
    Co-Authors: Michael Melese Woldeyohannis, Laurent Besacier, Million Meshesha
    Abstract:

    Speech translation research for the major languages like English, Japanese and Spanish has been conducted since the 1980’s. But no attempt were made in speech translation to/from the under-resourced language like Amharic. These activities suffered from the lack of Amharic speech and Amharic-English text corpus suited for the development of speech translation between the two languages. In this paper, therefore, an attempt has been made to collect, translate and record speech data from resourced language (English) to under-resourced language (Amharic) taking a Basic Traveler Expression Corpus (BTEC) as domain. Since there is no any Amharic text and speech corpus readily available for speech translation purposes, first, 7.43 h of Amharic read-speech has been prepared from 8,112 sentences, and second, 19,972 parallel Amharic-English corpus has been prepared taking tourism as an application domain. The Amharic speech data is recorded using smart-phone based application tool, LIG-Aikuma under a normal working environment. With the availability of such standard speech and text corpus, researcher will find a ground to further explore speech translation to/from under resourced languages.

  • Amharic english speech translation in tourism domain
    Empirical Methods in Natural Language Processing, 2017
    Co-Authors: Michael Melese, Laurent Besacier, Million Meshesha
    Abstract:

    This paper describes speech translation from Amharic-to-English, particularly Automatic Speech Recognition (ASR) with post-editing feature and Amharic-English Statistical Machine Translation (SMT). ASR experiment is conducted using morpheme language model (LM) and phoneme acoustic model(AM). Likewise,SMT conducted using word and morpheme as unit. Morpheme based translation shows a 6.29 BLEU score at a 76.4% of recognition accuracy while word based translation shows a 12.83 BLEU score using 77.4% word recognition accuracy. Further, after post-edit on Amharic ASR using corpus based n-gram, the word recognition accuracy increased by 1.42%. Since post-edit approach reduces error propagation, the word based translation accuracy improved by 0.25 (1.95%) BLEU score. We are now working towards further improving propagated errors through different algorithms at each unit of speech translation cascading component.

  • Amharic Speech Recognition for Speech Translation
    2016
    Co-Authors: Michael Melese, Laurent Besacier, Million Meshesha
    Abstract:

    The state-of-the-art speech translation can be seen as a cascade of Automatic Speech Recognition, Statistical Machine Translation and Text-To-Speech synthesis. In this study an attempt is made to experiment on Amharic speech recognition for Amharic-English speech translation in tourism domain. Since there is no Amharic speech corpus, we developed a read-speech corpus of 7.43hr in tourism domain. The Amharic speech corpus has been recorded after translating standard Basic Traveler Expression Corpus (BTEC) under a normal working environment. In our ASR experiments phoneme and syllable units are used for acoustic models, while morpheme and word are used for language models. Encouraging ASR results are achieved using morpheme-based language models and phoneme-based acoustic models with a recognition accuracy result of 89.1%, 80.9%, 80.6%, and 49.3% at character, morph, word and sentence level respectively. We are now working towards designing Amharic-English speech translation through cascading components under different error correction algorithms.

Laurent Besacier - One of the best experts on this subject based on the ideXlab platform.

  • a corpus for Amharic english speech translation the case of tourism domain
    International Conference on Information and Communication Technology, 2017
    Co-Authors: Michael Melese Woldeyohannis, Laurent Besacier, Million Meshesha
    Abstract:

    Speech translation research for the major languages like English, Japanese and Spanish has been conducted since the 1980’s. But no attempt were made in speech translation to/from the under-resourced language like Amharic. These activities suffered from the lack of Amharic speech and Amharic-English text corpus suited for the development of speech translation between the two languages. In this paper, therefore, an attempt has been made to collect, translate and record speech data from resourced language (English) to under-resourced language (Amharic) taking a Basic Traveler Expression Corpus (BTEC) as domain. Since there is no any Amharic text and speech corpus readily available for speech translation purposes, first, 7.43 h of Amharic read-speech has been prepared from 8,112 sentences, and second, 19,972 parallel Amharic-English corpus has been prepared taking tourism as an application domain. The Amharic speech data is recorded using smart-phone based application tool, LIG-Aikuma under a normal working environment. With the availability of such standard speech and text corpus, researcher will find a ground to further explore speech translation to/from under resourced languages.

  • Amharic english speech translation in tourism domain
    Empirical Methods in Natural Language Processing, 2017
    Co-Authors: Michael Melese, Laurent Besacier, Million Meshesha
    Abstract:

    This paper describes speech translation from Amharic-to-English, particularly Automatic Speech Recognition (ASR) with post-editing feature and Amharic-English Statistical Machine Translation (SMT). ASR experiment is conducted using morpheme language model (LM) and phoneme acoustic model(AM). Likewise,SMT conducted using word and morpheme as unit. Morpheme based translation shows a 6.29 BLEU score at a 76.4% of recognition accuracy while word based translation shows a 12.83 BLEU score using 77.4% word recognition accuracy. Further, after post-edit on Amharic ASR using corpus based n-gram, the word recognition accuracy increased by 1.42%. Since post-edit approach reduces error propagation, the word based translation accuracy improved by 0.25 (1.95%) BLEU score. We are now working towards further improving propagated errors through different algorithms at each unit of speech translation cascading component.

  • Amharic Speech Recognition for Speech Translation
    2016
    Co-Authors: Michael Melese, Laurent Besacier, Million Meshesha
    Abstract:

    The state-of-the-art speech translation can be seen as a cascade of Automatic Speech Recognition, Statistical Machine Translation and Text-To-Speech synthesis. In this study an attempt is made to experiment on Amharic speech recognition for Amharic-English speech translation in tourism domain. Since there is no Amharic speech corpus, we developed a read-speech corpus of 7.43hr in tourism domain. The Amharic speech corpus has been recorded after translating standard Basic Traveler Expression Corpus (BTEC) under a normal working environment. In our ASR experiments phoneme and syllable units are used for acoustic models, while morpheme and word are used for language models. Encouraging ASR results are achieved using morpheme-based language models and phoneme-based acoustic models with a recognition accuracy result of 89.1%, 80.9%, 80.6%, and 49.3% at character, morph, word and sentence level respectively. We are now working towards designing Amharic-English speech translation through cascading components under different error correction algorithms.

  • phoneme based english Amharic statistical machine translation
    AFRICON, 2015
    Co-Authors: Mulu Gebreegziabher Teshome, Laurent Besacier, Girma Taye, Dereje Teferi
    Abstract:

    This research considers the application of Statistical method to automatic Machine Translation (MT) from English to Amharic. The research focuses on improving the translation quality by applying phonemic transcription on the target side, which is Amharic. Accordingly, the BLEU score results for the phoneme-based EASMT system is 37.53% a gain of 2.21 BLEU point from another baseline phrase-based EASMT with a BLEU score result of 35.32%. This clearly shows that phoneme-based translation outperforms the baseline system.

  • preliminary experiments on english Amharic statistical machine translation
    SLTU, 2012
    Co-Authors: Mulu Gebreegziabher Teshome, Laurent Besacier
    Abstract:

    This paper discusses the preliminary experiment conducted to translate from English to Amharic using the Statistical Machine Translation (EASMT) approach. The experiment on the EASMT system is being conducted on training corpus of both languages based on expressions that are found in parallel documents. The experiment involves collecting of a total of 632 Parliamentary corpora of which 115 have been used in the experiment. The corpus coverage is 15 years from Aug 21, 1995 to July 16, 2010. Each document contains data, which are translations of each other. The experiment has been conducted using 18,432 English-Amharic sentence pairs extracted from these corpora in order to measure the accuracy of the translation system. Accordingly, the baseline phrase-based BLEU score result is 35.32%. A 0.34% increase in BLEU has been achieved by applying morpheme segmentation to the tokens of the Amharic output result and the reference of the baseline system. The increase is 0.92% when compared with the same segmented reference between the baseline and the segmented system.

Yaregal Assabie - One of the best experts on this subject based on the ideXlab platform.

  • Amharic document representation for adhoc retrieval
    International Joint Conference on Knowledge Discovery Knowledge Engineering and Knowledge Management, 2020
    Co-Authors: Tilahun Yeshambel, Josiane Mothe, Yaregal Assabie
    Abstract:

    Amharic is the official language of the government of Ethiopia currently having an estimated population of over 110 million. Like other Semitic languages, Amharic is characterized by complex morphology where thousands of words are generated from a single root form through inflection and derivation. This has made the development of tools for Amharic natural language processing a non-trivial task. Amharic adhoc retrieval faces difficulties due to the complex morphological structure of the language. In this paper, the impact of morphological features on the representation of Amharic documents and queries for adhoc retrieval is investigated. We analyze the effects of stem-based and root-based approaches on Amharic adhoc retrieval effectiveness. Various experiments are conducted on TREC-like Amharic information retrieval test collection using standard evaluation framework and measures. The findings show that a root-based approach outperforms the conventional stem-based approachthat prevails in many other languages.

  • 2airtc the Amharic adhoc information retrieval test collection
    Cross-Language Evaluation Forum, 2020
    Co-Authors: Tilahun Yeshambel, Josiane Mothe, Yaregal Assabie
    Abstract:

    Evaluation is highly important for designing, developing, and maintaining information retrieval (IR) systems. The IR community has developed shared tasks where evaluation framework, evaluation measures and test collections have been developed for different languages. Although Amharic is the official language of Ethiopia currently having an estimated population of over 110 million, it is one of the under-resourced languages and there is no Amharic adhoc IR test collection to date. In this paper, we promote the monolingual Amharic IR test collection that we build for the IR community. Following the framework of Cranfield project and TREC, the collection that we named 2AIRTC consists of 12,583 documents, 240 topics and the corresponding relevance judgments.

  • large vocabulary read speech corpora for four ethiopian languages Amharic tigrigna oromo and wolaytta
    Language Resources and Evaluation, 2020
    Co-Authors: Solomon Teferra Abate, Yaregal Assabie, Michael Melese, Wondwossen Mulugeta, Martha Yifiru Tachbelie, Hafte Abera, Tewodros Gebreselassie, Million Meshesha Beyene, Solomon Atinafu, Binyam Ephrem Seyoum
    Abstract:

    Automatic Speech Recognition (ASR) is one of the most important technologies to support spoken communication in modern life. However, its development benefits from large speech corpus. The development of such a corpus is expensive and most of the human languages, including the Ethiopian languages, do not have such resources. To address this problem, we have developed four large (about 22 hours) speech corpora for four Ethiopian languages: Amharic, Tigrigna, Oromo and Wolaytta. To assess usability of the corpora for (the purpose of) speech processing, we have developed ASR systems for each language. In this paper, we present the corpora and the baseline ASR systems we have developed. We have achieved word error rates (WERs) of 37.65%, 31.03%, 38.02%, 33.89% for Amharic, Tigrigna, Oromo and Wolaytta, respectively. This results show that the corpora are suitable for further investigation towards the development of ASR systems. Thus, the research community can use the corpora to further improve speech processing systems. From our results, it is clear that the collection of text corpora to train strong language models for all of the languages is still required, especially for Oromo and Wolaytta.

  • Amharic sentence parsing using base phrase chunking
    International Conference on Computational Linguistics, 2014
    Co-Authors: Abeba Ibrahim, Yaregal Assabie
    Abstract:

    Parsing plays a significant role in many natural language processing NLP applications as their efficiency relies on having an effective parser. This paper presents Amharic sentence parser developed using base phrase chunker that groups syntactically correlated words at different levels. We use HMM to chunk base phrases where incorrectly chunked phrases are pruned with rules. The task of parsing is then performed by taking chunk results as inputs. Bottom-up approach with transformation algorithm is used to transform the chunker to the parser. Corpus from Amharic news outlets and books was collected for training and testing. The training and testing datasets were prepared using the 10-fold cross validation technique. Test results on the test data showed an average parsing accuracy of 93.75%.

  • hierarchical Amharic base phrase chunking using hmm with error pruning
    Language and Technology Conference, 2013
    Co-Authors: Abeba Ibrahim, Yaregal Assabie
    Abstract:

    Segmentation of a text into non-overlapping syntactic units (chunks) has become an essential component of many applications of natural language processing. This paper presents Amharic base phrase chunker that groups syntactically correlated words at different levels using HMM. Rules are used to correct chunk phrases incorrectly chunked by the HMM. For the identification of the boundary of the phrases IOB2 chunk specification is selected and used in this work. To test the performance of the system, corpus was collected from Amharic news outlets and books. The training and testing datasets were prepared using the 10-fold cross validation technique. Test results on the corpus showed an average accuracy of 85.31 % before applying the rule for error correction and an average accuracy of 93.75 % after applying rules.

Michael Melese - One of the best experts on this subject based on the ideXlab platform.

  • large vocabulary read speech corpora for four ethiopian languages Amharic tigrigna oromo and wolaytta
    Language Resources and Evaluation, 2020
    Co-Authors: Solomon Teferra Abate, Yaregal Assabie, Michael Melese, Wondwossen Mulugeta, Martha Yifiru Tachbelie, Hafte Abera, Tewodros Gebreselassie, Million Meshesha Beyene, Solomon Atinafu, Binyam Ephrem Seyoum
    Abstract:

    Automatic Speech Recognition (ASR) is one of the most important technologies to support spoken communication in modern life. However, its development benefits from large speech corpus. The development of such a corpus is expensive and most of the human languages, including the Ethiopian languages, do not have such resources. To address this problem, we have developed four large (about 22 hours) speech corpora for four Ethiopian languages: Amharic, Tigrigna, Oromo and Wolaytta. To assess usability of the corpora for (the purpose of) speech processing, we have developed ASR systems for each language. In this paper, we present the corpora and the baseline ASR systems we have developed. We have achieved word error rates (WERs) of 37.65%, 31.03%, 38.02%, 33.89% for Amharic, Tigrigna, Oromo and Wolaytta, respectively. This results show that the corpora are suitable for further investigation towards the development of ASR systems. Thus, the research community can use the corpora to further improve speech processing systems. From our results, it is clear that the collection of text corpora to train strong language models for all of the languages is still required, especially for Oromo and Wolaytta.

  • Amharic english speech translation in tourism domain
    Empirical Methods in Natural Language Processing, 2017
    Co-Authors: Michael Melese, Laurent Besacier, Million Meshesha
    Abstract:

    This paper describes speech translation from Amharic-to-English, particularly Automatic Speech Recognition (ASR) with post-editing feature and Amharic-English Statistical Machine Translation (SMT). ASR experiment is conducted using morpheme language model (LM) and phoneme acoustic model(AM). Likewise,SMT conducted using word and morpheme as unit. Morpheme based translation shows a 6.29 BLEU score at a 76.4% of recognition accuracy while word based translation shows a 12.83 BLEU score using 77.4% word recognition accuracy. Further, after post-edit on Amharic ASR using corpus based n-gram, the word recognition accuracy increased by 1.42%. Since post-edit approach reduces error propagation, the word based translation accuracy improved by 0.25 (1.95%) BLEU score. We are now working towards further improving propagated errors through different algorithms at each unit of speech translation cascading component.

  • Amharic Speech Recognition for Speech Translation
    2016
    Co-Authors: Michael Melese, Laurent Besacier, Million Meshesha
    Abstract:

    The state-of-the-art speech translation can be seen as a cascade of Automatic Speech Recognition, Statistical Machine Translation and Text-To-Speech synthesis. In this study an attempt is made to experiment on Amharic speech recognition for Amharic-English speech translation in tourism domain. Since there is no Amharic speech corpus, we developed a read-speech corpus of 7.43hr in tourism domain. The Amharic speech corpus has been recorded after translating standard Basic Traveler Expression Corpus (BTEC) under a normal working environment. In our ASR experiments phoneme and syllable units are used for acoustic models, while morpheme and word are used for language models. Encouraging ASR results are achieved using morpheme-based language models and phoneme-based acoustic models with a recognition accuracy result of 89.1%, 80.9%, 80.6%, and 49.3% at character, morph, word and sentence level respectively. We are now working towards designing Amharic-English speech translation through cascading components under different error correction algorithms.

Solomon Teferra Abate - One of the best experts on this subject based on the ideXlab platform.

  • large vocabulary read speech corpora for four ethiopian languages Amharic tigrigna oromo and wolaytta
    Language Resources and Evaluation, 2020
    Co-Authors: Solomon Teferra Abate, Yaregal Assabie, Michael Melese, Wondwossen Mulugeta, Martha Yifiru Tachbelie, Hafte Abera, Tewodros Gebreselassie, Million Meshesha Beyene, Solomon Atinafu, Binyam Ephrem Seyoum
    Abstract:

    Automatic Speech Recognition (ASR) is one of the most important technologies to support spoken communication in modern life. However, its development benefits from large speech corpus. The development of such a corpus is expensive and most of the human languages, including the Ethiopian languages, do not have such resources. To address this problem, we have developed four large (about 22 hours) speech corpora for four Ethiopian languages: Amharic, Tigrigna, Oromo and Wolaytta. To assess usability of the corpora for (the purpose of) speech processing, we have developed ASR systems for each language. In this paper, we present the corpora and the baseline ASR systems we have developed. We have achieved word error rates (WERs) of 37.65%, 31.03%, 38.02%, 33.89% for Amharic, Tigrigna, Oromo and Wolaytta, respectively. This results show that the corpora are suitable for further investigation towards the development of ASR systems. Thus, the research community can use the corpora to further improve speech processing systems. From our results, it is clear that the collection of text corpora to train strong language models for all of the languages is still required, especially for Oromo and Wolaytta.

  • effect of language resources on automatic speech recognition for Amharic
    AFRICON, 2015
    Co-Authors: Martha Yifiru Tachbelie, Solomon Teferra Abate
    Abstract:

    This paper presents our investigation of the effect of language resources on the performance of Amharic speech recognition. We have used language model training text of different sizes and seen the effect on word error rate (WER) reduction. Moreover, we have investigated the effect of handling language issues (germination, epenthetic vowel insertion and glottal stop consonant pronunciation) on the performance of speech recognition systems using data-driven phone-level transcriptions. The results of our experiments show that only slight reduction in WER can be obtained by increasing language model training text. However, proper transcription of gemination, the epenthetic vowel and the glottal stop consonant did not bring performance improvement for Amharic speech recognition. This can be attributed to the larger number of phone HMM acoustic models (62 compared to 37 phone set of the grapheme-based phone-level transcriptions) trained with a small (5 hrs) training speech.

  • quality assessment of crowdsourcing transcriptions for african languages
    Conference of the International Speech Communication Association, 2011
    Co-Authors: Hadrien Gelas, Laurent Besacier, Solomon Teferra Abate, Francois Pellegrino
    Abstract:

    We evaluate the quality of speech transcriptions acquired by crowdsourcing to develop ASR acoustic models (AM) for under-resourced languages. We have developed AMs using reference (REF) transcriptions and transcriptions from crowdsourcing (TRK) for Swahili and Amharic. While the Amharic transcription was much slower than that of Swahili to complete, the speech recognition systems developed using REF and TRK transcriptions have almost similar (40.1 vs 39.6 for Amharic and 38.0 vs 38.5 for Swahili) word recognition error rate. Moreover, the character level disagreement rates between REF and TRK are only 3.3% and 6.1% for Amharic and Swahili, respectively. We conclude that it is possible to acquire quality transcriptions from the crowd for under-resourced languages using Amazon’s Mechanical Turk. Recognizing such a great potential of it, we recommend some legal and ethical issues to consider.

  • part of speech tagging for under resourced and morphologically rich languages the case of Amharic
    Human Language Technologies for Development (HLTD), 2011
    Co-Authors: Martha Yifiru Tachbelie, Solomon Teferra Abate, Laurent Besacier
    Abstract:

    This paper presents part-of-speech (POS) tagging experiments conducted to identify the best method for under-resourced and morphologically rich languages. The experiments have been conducted using different tagging strategies and different training data sizes for Amharic. Experiments on word segmentation and tag hypotheses combination have also been conducted to improve tagging accuracy. The results showed that methods like MBT are good for under-resourced languages. Moreover, segmenting words composed of morphemes of different POS tags and tag hypotheses combination are promising directions to improve tagging performance for under-resourced and morphologically rich languages.

  • syllable based speech recognition for Amharic
    Meeting of the Association for Computational Linguistics, 2007
    Co-Authors: Solomon Teferra Abate, Wolfgang Menzel
    Abstract:

    Amharic is the Semitic language that has the second large number of speakers after Arabic (Hayward and Richard 1999). Its writing system is syllabic with Consonant-Vowel (CV) syllable structure. Amharic orthography has more or less a one to one correspondence with syllabic sounds. We have used this feature of Amharic to develop a CV syllable-based speech recognizer, using Hidden Markov Modeling (HMM), and achieved 90.43% word recognition accuracy.