Turdybek Kurmetkan
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Turdybek Kurmetkan has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Orken Mamyrbayev
Institute of Information and Computational Technologies, Almaty 050010, Kazakhstan
6 shared publications - Dina Oralbekova
Institute of Information and Computational Technologies, Almaty, Kazakhstan
3 shared publications - Nurdaulet Zhumazhan
U.Joldasbekov Institute of Mechanics and Engineering
3 shared publications - Adem Tekerek
Computer Engineering Department, Technology Faculty, Gazi University, Ankara, Turkey
1 shared publication - Ainur Toleu
National Scientific and Practical Center “Til–Kazyna” Named After Shaysultan Shayakhmetov, 010000 Astana, Kazakhstan
1 shared publication - Dinara Mussayeva
Institute of Economics of the Committee of Science, Almaty 050010, Kazakhstan
1 shared publication
Latest Publications
- Open Access
- 6 August 2026
- Published by MDPI AG in Information
- Vol. 17 (8) , 756
Developing a Kazakh Audio–Visual Multimodal Speech Recognition Model Based on Hierarchical and Cross-Modal Attention
This study presents an audio–visual speech recognition (AVSR) model for Kazakh that jointly exploits audio and visual channels. The study introduces QazAVSR, a 57 h dataset collected from 271 speakers, and extracts synchronized audio signals and lip-region video sequences using FFmpeg 7.0, Dlib 19.24, and OpenCV 4.9.0. The proposed architecture uses the self-supervised HuBERT_BASE model in the audio branch and an ImageNet-pretrained ViT-B/16 model in the visual branch. Audio and visual representations are fused by a three-layer BiModalHformer block, where intra- and cross-attention operations are performed at each level. Extensive experimental validation, supplemented by rigorous paired bootstrap resampling significance tests, demonstrates that the full multimodal BiModalHformer model achieves a highly robust average character error rate (CER) of 31.2% and a Word Error Rate (WER) of 43.1%. These results significantly outperform traditional audio-only, video-only, and standard representation-level fusion baselines. Furthermore, comparisons against powerful external baseline architectures—including Whisper-Small and AV-HuBERT configurations rigorously adapted for the Kazakh language—statistically validate the architectural efficacy of the BiModalHformer framework. Additional systematic evaluations utilizing extended metrics such as the Match Error Rate (MER), word information preserved (WIP), and the Multimodal Synergy Index (MSI) confirm that the full audio–visual configuration preserves lexical information significantly more effectively. Finally, extensive noise perturbation experiments confirm that the multimodal architecture exhibits superior structural robustness to complex acoustic distortions, including environmental noise, synthetic room reverberation, and overlapping speech topologies.
- 6 August 2026
- Published by MDPI AG in Information
- Vol. 17 (8) , 756
Open AccessDeveloping a Kazakh Audio–Visual Multimodal Speech Recognition Model Based on Hierarchical and Cross-Modal Attention
This study presents an audio–visual speech recognition (AVSR) model for Kazakh that jointly exploits audio and visual channels. The study introduces QazAVSR, a 57 h dataset collected from 271 speakers, and extracts synchronized audio signals and lip-region video sequences using FFmpeg 7.0, Dlib 19.24, and OpenCV 4.9.0. The proposed architecture uses the self-supervised HuBERT_BASE model in the audio branch and an ImageNet-pretrained ViT-B/16 model in the visual branch. Audio and visual representations are fused by a three-layer BiModalHformer block, where intra- and cross-attention operations are performed at each level. Extensive experimental validation, supplemented by rigorous paired bootstrap resampling significance tests, demonstrates that the full multimodal BiModalHformer model achieves a highly robust average character error rate (CER) of 31.2% and a Word Error Rate (WER) of 43.1%. These results significantly outperform traditional audio-only, video-only, and standard representation-level fusion baselines. Furthermore, comparisons against powerful external baseline architectures—including Whisper-Small and AV-HuBERT configurations rigorously adapted for the Kazakh language—statistically validate the architectural efficacy of the BiModalHformer framework. Additional systematic evaluations utilizing extended metrics such as the Match Error Rate (MER), word information preserved (WIP), and the Multimodal Synergy Index (MSI) confirm that the full audio–visual configuration preserves lexical information significantly more effectively. Finally, extensive noise perturbation experiments confirm that the multimodal architecture exhibits superior structural robustness to complex acoustic distortions, including environmental noise, synthetic room reverberation, and overlapping speech topologies.
- Open Access
- 22 April 2026
- Published by MDPI AG in Information
- Vol. 17 (5) , 398
Enhanced Sentiment Analysis of E-Commerce Product Reviews Using Luong Attention-Based Bi-LSTM
The rapid growth of e-commerce has highlighted the critical need for efficient customer review sentiment analysis, yet natural language complexities like sarcasm and mixed sentiments remain challenging. To address these ambiguities, this study proposes a novel sentiment analysis architecture. The methodology integrates a bidirectional Long Short-Term Memory (Bi-LSTM) network with a Luong Attention mechanism. The Bi-LSTM component models the sequential and bidirectional context of the text, while the Luong Attention mechanism isolates and emphasizes the most significant parts of the reviews for precise sentiment detection. The proposed hybrid model demonstrates exceptional performance compared to traditional methods, achieving an accuracy of 96.67%, a precision of 96.83%, and a recall of 96.67%, alongside relatively low overfitting. Ultimately, the findings confirm that this architecture effectively manages ambiguous language and is highly capable of large-scale, real-time sentiment analysis, offering robust analytical tools for shaping e-commerce marketing strategies.
- 22 April 2026
- Published by MDPI AG in Information
- Vol. 17 (5) , 398
Open AccessEnhanced Sentiment Analysis of E-Commerce Product Reviews Using Luong Attention-Based Bi-LSTM
The rapid growth of e-commerce has highlighted the critical need for efficient customer review sentiment analysis, yet natural language complexities like sarcasm and mixed sentiments remain challenging. To address these ambiguities, this study proposes a novel sentiment analysis architecture. The methodology integrates a bidirectional Long Short-Term Memory (Bi-LSTM) network with a Luong Attention mechanism. The Bi-LSTM component models the sequential and bidirectional context of the text, while the Luong Attention mechanism isolates and emphasizes the most significant parts of the reviews for precise sentiment detection. The proposed hybrid model demonstrates exceptional performance compared to traditional methods, achieving an accuracy of 96.67%, a precision of 96.83%, and a recall of 96.67%, alongside relatively low overfitting. Ultimately, the findings confirm that this architecture effectively manages ambiguous language and is highly capable of large-scale, real-time sentiment analysis, offering robust analytical tools for shaping e-commerce marketing strategies.
- 30 December 2024
- Published by National Engineering Academy of the Republic of Kazakhstan in Bulletin of the National Engineering Academy of the Republic of Kazakhstan
- Vol. 94 (4) , 290-301
Analysis of the use of the hiformer model for kazakh speech recognition
This article presents an overview of automatic speech recognition (ASR) technologies and describes the use of an advanced version of the Transformer model, the Hiformer model, in Kazakh speech recognition. A literature review of Kazakh speech recognition systems was made. The structure of the Hiformer model is described and how it can be used in different parts of the structure of an advanced attention mechanism (AED) (encoder, decoder, cross-coder attention). An experiment was carried out on the execution of tasks of recognition of Kazakh speech using the Hiformer model. This study also details an experiment conducted to assess the efficacy of the HiFormer model in recognizing Kazakh speech. The experimental results clearly demonstrate the superiority and enhanced efficiency of the HiFormer model over its predecessors, the Transformer and Conformer models, in handling the complexities of Kazakh speech recognition. This finding underscores the potential of the HiFormer model in advancing ASR technology for languages with unique linguistic characteristics. In recognizing Kazakh speech, the use of the Hiformer model reduced the word error rate (WER) by 2.6 % and the character error rate (CER) by 1.4% compared to the Transformer model.
- 30 December 2024
- Published by National Engineering Academy of the Republic of Kazakhstan in Bulletin of the National Engineering Academy of the Republic of Kazakhstan
- Vol. 94 (4) , 290-301
Analysis of the use of the hiformer model for kazakh speech recognition
This article presents an overview of automatic speech recognition (ASR) technologies and describes the use of an advanced version of the Transformer model, the Hiformer model, in Kazakh speech recognition. A literature review of Kazakh speech recognition systems was made. The structure of the Hiformer model is described and how it can be used in different parts of the structure of an advanced attention mechanism (AED) (encoder, decoder, cross-coder attention). An experiment was carried out on the execution of tasks of recognition of Kazakh speech using the Hiformer model. This study also details an experiment conducted to assess the efficacy of the HiFormer model in recognizing Kazakh speech. The experimental results clearly demonstrate the superiority and enhanced efficiency of the HiFormer model over its predecessors, the Transformer and Conformer models, in handling the complexities of Kazakh speech recognition. This finding underscores the potential of the HiFormer model in advancing ASR technology for languages with unique linguistic characteristics. In recognizing Kazakh speech, the use of the Hiformer model reduced the word error rate (WER) by 2.6 % and the character error rate (CER) by 1.4% compared to the Transformer model.
- Open Access
- 3 December 2024
- Published by Springer Nature
- p. C1
Correction to: A Study of Kazakh Speech Recognition in Hiformer Model
- 3 December 2024
- Published by Springer Nature
- p. C1
Open AccessCorrection to: A Study of Kazakh Speech Recognition in Hiformer Model
- 9 September 2024
- Published by Springer Nature
- p. 273-283
Recent Methods and Algorithms in Speech Segmentation Tasks
- 9 September 2024
- Published by Springer Nature
- p. 273-283
Recent Methods and Algorithms in Speech Segmentation Tasks
- 13 August 2024
- Published by Springer Nature
- p. 330-340
A Study of Kazakh Speech Recognition in Hiformer Model
- 13 August 2024
- Published by Springer Nature
- p. 330-340
A Study of Kazakh Speech Recognition in Hiformer Model