College of Computing and Informatics

URI for this communityhttps://rps.wku.edu.et/handle/987654321/2333

College of Computing and Informatics

Browse

Filters

2023 - 20252

Settings

Has files: YesSubject: search.filters.subject.Deep learning

Search Results

Now showing 1 - 2 of 2
  • No Thumbnail Available
    Item
    END-TO-END SPEECH RECOGNITION FOR GURAGIGNA LANGUAGE USING DEEP LEARNING TECHNIQUES
    (Wolkite University, 2025-10-05) ABDO NESRU EBRAHIM
    Speech recognition entails converting long sequences of acoustic features into shorter sequences of discrete symbols, such as words or phonemes. This process is complicated by varying sequence lengths and uncertainty in output symbol locations, making traditional classifiers impractical. Current automated systems struggle with speaker-independent continuous speech, particularly in low-resource languages like Guragigna, where the Cheha dialect poses additional challenges due to its purely spoken nature and lack of a rigid grammatical structure. To address these issues, this research develops an end-to-end speech recognition model utilizing deep learning techniques, specifically a hybrid CNN-BIGRU architecture combined with CTC and attention mechanisms. This approach aims to enhance alignment and robustness in noisy environments. To train and test the model, a text and speech corpus was created by compiling dataset from different sources like in Wolkite FM, the Old and New Testaments. Experimental results indicate that the CNN-BIGRU model achieves a Word Error Rate (WER) of 2.5%, showcasing improved generalization capabilities. Additionally, four recurrent neural network models LSTM, Bilstm, GRU, and BIGRU were evaluated, each configured with 1024 hidden units and optimized using the Adam optimizer over 50 epochs. The BIGRU model outperformed the others, achieving an accuracy of 97.50%, while the LSTM, Bilstm, and GRU models achieved maximum accuracies of 95.99%, 96.92%, and 96.25%, respectively. The successful implementation of this end-to-end speech recognition system significantly advances communication technologies for low-resource languages, enhancing accessibility for diverse linguistic communities. The findings underscore the effectiveness of deep learning methods in improving speech recognition performance in challenging linguistic contexts.
  • No Thumbnail Available
    Item
    YEMSA TO AMHARIC MACHINE TRANSLATION USINGDEEP LEARNING TECHNIQUES
    (Wolkite University, 2023-04-01) TEMESGEN HABTAMU ESHETU
    In today's globalized world, the barriers of distance and language have been greatly diminished, transforming our world into a closely interconnected global community. As a consequence, human languages have taken on an international character, enabling effective communication across borders. Traditionally, human translation is costly and inconvenient; several kinds of research are currently conducted to resolve this problem with machine translation techniques. |So, it is automatic, which means it translates one language to another using a computer software system. In this study, Yemsa to Amharic machine translation and vice versa are used by deep learning techniques. Due to increased speaker numbers, to address the issue of endangered Yemsa language and enhance the language's content on the World Wide Web. A numberof indigenous knowledge medicines called Samo Heta and other traditional and religious names are found in the Yemsa language. We utilized the current STOA method of deep learning. The work was executed using a seq-to-seq encoder-decoder architecture. The proposed study was conducted experiments on LSTM, Bi-LSTM, LSTM with attention, GRU and transformer models. We collected a dataset of about 6,000 parallel sentences with 11690 and 12491 vocabularies. In order to translate textinto sentence sequence, we applied the preprocessing technique and used Morfessor tools. The proposed studies utilize the 80/20 splitting technique for dividing the dataset into training and testing sets. The next step is training and testing models on a corresponding with training and testing dataset. The experiment was conducted on LSTM, Bi-LSTM, LSTM+ attention, GRU and Transformer models. Among those models, the transformer model outperforms other models by 99.4% accuracy, 0.0113 loss. And BLEU scores of 9.7 and 9.8 from Yemsa to Amharic and Amharic to Yemsa respectively. The primary limitation of the investigation is the insufficient availability of a substantial dataset to conduct comprehensive experimentation. As a result, there is a necessity to generate parallel corpora in order to conduct comparable research. Finally, the findings of the study show that utilizing deep learning techniques, particularly the transformer model, can significantly improve Yemsa to Amharic machine translation accuracy and BLEU scores.