This study examines how ten different audio features, including MFCC, mel-spectrogram, chroma, and spectral contrast etc., influence speaker change detection (SCD) performance. The analysis is conducted using two unsupervised methods: Bayesian information criterion with Gaussian mixture model (BIC-GMM), a model-based approach, and Kullback-Leibler divergence with Gaussian Mixture Model (KL-GMM), a metric-based approach. Evaluation involved statistical analysis of feature changes in relation to speaker changes (vice versa), supported by comprehensive experimental validation. Experimental results show MFCC as the most effective feature, demonstrating consistently good performance across both methods. Features such as zero crossing rate, chroma, and spectral contrast also showed notable effectiveness within the BIC-GMM framework, while mel-spectrogram consistently ranked as the least influential feature in both approaches. Further analysis revealed that BIC-GMM exhibits greater stability in managing variations in feature performance, whereas KL-GMM is more sensitive to threshold optimization. Nevertheless, KL-GMM achieved competitive results when paired with specific features, such as MFCC and zero crossing rate. These findings offer valuable insights into the impact of feature selection on unsupervised SCD, providing guidance for the development of more robust and accurate algorithms for practical applications.
Kazakh is an agglutinative language, where the sequential attachment of morphemes forms words, each bearing specific grammatical information. The complexity of this morphological structure presents significant challenges for computational linguistics, particularly in natural language processing tasks such as part-of-speech tagging, syntactic parsing, and machine translation. This paper provides a comprehensive review of the most widely used machine learning approaches applied to Kazakh morphology. It compares classical machine learning techniques, such as hidden Markov models and conditional random fields, along with rule-based methods, and recent advances in deep learning, including recurrent neural networks, long short-term memory networks, and transformer-based architectures like BERT. The paper also explores the applicability of data-driven approaches and reviews existing research on Kazakh morphology, offering a comparative analysis based on efficiency, scalability, and accuracy. Finally, future directions are discussed, focusing on expanding annotated datasets and developing models tailored for low-resource languages like Kazakh to improve their representation in natural language processing tasks.
This article presents an approach for spellchecking and autocorrection using web data for morphologically complex languages (in the case of Kazakh language), which can be considered an end-to-end approach that does not require any manually annotated word–error pairs. A sizable web of noisy data is crawled and used as a base to infer the knowledge of misspellings with their correct forms. Using the extracted corpus, a sub-string error model with a context model for morphologically complex languages are trained separately, then these two models are integrated with a regularization parameter. A sub-string alignment model is applied to extract symmetric and non-symmetric patterns in two sequences of word–error pairs. The model calculates the probability for symmetric and non-symmetric patterns of a given misspelling and its candidates to obtain a suggestion list. Based on the proposed method, a Kazakh Spellchecking and Autocorrection system is developed, which we refer to as QazSpell. Several experiments are conducted to evaluate the proposed approach from different angles. The results show that the proposed approach achieves a good outcome when only using the error model, and the performance is boosted after integrating the context model. In addition, the developed system, QazSpell, outperforms the commercial analogs in terms of overall accuracy.
This study investigates semantic text segmentation enhanced by large language model (LLM) embeddings. We assess how effectively embeddings capture semantic coherence and topic closure by integrating them into both classical clustering algorithms and a modified graph-based methods. In addition, we propose a simple magnetic clustering algorithm as a lightweight baseline. Experiments are conducted across multiple datasets and embedding models, with segmentation quality evaluated using the boundary segmentation metric. Results demonstrate that LLM embeddings improve segmentation accuracy, highlight dataset-specific difficulties, and reveal how contextual window size and embedding choice affect performance. These findings clarify the strengths and limitations of embedding-based approaches to segmentation and provide insights relevant to retrieval-augmented generation (RAG).
Mean Shift is a flexible, non-parametric clustering algorithm that identifies dense regions in data through gradient ascent on a kernel density estimate. Its ability to detect arbitrarily shaped clusters without requiring prior knowledge of the number of clusters makes it widely applicable across diverse domains. However, its quadratic computational complexity restricts its use on large or high-dimensional datasets. Numerous acceleration techniques, collectively referred to as Fast Mean Shift strategies, have been developed to address this limitation while preserving clustering quality. This paper presents a systematic theoretical analysis of these strategies, focusing on their computational impact, pairwise combinability, and mapping onto distinct stages of the Mean Shift pipeline. Acceleration methods are categorized into seed reduction, neighborhood search acceleration, adaptive bandwidth selection, kernel approximation, and parallelization, with their algorithmic roles examined in detail. A pairwise compatibility matrix is proposed to characterize synergistic and conflicting interactions among strategies. Building on this analysis, we introduce a decision framework for selecting suitable acceleration strategies based on dataset characteristics and computational constraints. This framework, together with the taxonomy, combinability analysis, and scenario-based recommendations, establishes a rigorous foundation for understanding and systematically applying Fast Mean Shift methods.