Використання машинного навчання у задачах класифікації звуків
Loading...
Date
2019
Authors
Глибовець, Микола
Жиркова, Анастасія
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
У роботі розглянуто особливості використання методів машинного навчання (МН) для класифікації звукової інформації на прикладі розв’язку задачі класифікації міських звуків (МЗ). Дослідження з аналізу міських акустичних середовищ є досить обмеженими. Більше того, у цих дослідженнях основна увага фокусується на класифікації місць, які характеризують певні звуки, наприклад, парку, вулиці, на відміну від ідентифікації джерел звуку в них, таких як автомобільний сигнал, постріл тощо. Тому класифікація МЗ – досить актуальна проблема, що потребує вирішення. Метою роботи є висвітлення побудови оптимальних моделей МН для задачі коректної класифікації МЗ.
The article describes usaging circumstances of Machine Learning (ML) methods for classifying auditory information . The problem was studied by using an example of urban sounds classification. Machine Learning is a great possibility to process data presented differently, including text, images, and sounds. Natural language processing requires text analysis. Pattern recognition involves interaction with images. And, naturally, many ML tasks require working with numbers, e.g. prediction of the price for houses, or benign or malignant tumor recognition based on numerical characteristics of the tumor. As for sounds, most of researche is about music recognition and, in general, the related stuff. So, sound processing is a relevant topic for ML problems. Environmental sounds classification is a field which attracts specialists from various spheres. Multimedia sensor networks and big variety of multimedia content, which describes urban scenes, is a huge part of it. Urban sounds classification is an important problem which needs to be solved. Many researchers focus on locations (parks, streets, etc.) instead of sources (car horn, shoot, etc.). So, urban sounds classification is a reasonable, urgent problem. Development of best-fit Machine Learning models for correct classification of urban sounds requires dealing with the following problems: sound representation as a set of numerical data considering its features for a possibility of ML algorithms to work with sound; data processing for optimization of ML models efficiency; using methods of ML models optimization for presenting the best parameters of each model; train models on training data; prediction of urban sound classes and checking the accuracy of these predictions on test data; neural networks development for solving the urban sounds classification problem.
The article describes usaging circumstances of Machine Learning (ML) methods for classifying auditory information . The problem was studied by using an example of urban sounds classification. Machine Learning is a great possibility to process data presented differently, including text, images, and sounds. Natural language processing requires text analysis. Pattern recognition involves interaction with images. And, naturally, many ML tasks require working with numbers, e.g. prediction of the price for houses, or benign or malignant tumor recognition based on numerical characteristics of the tumor. As for sounds, most of researche is about music recognition and, in general, the related stuff. So, sound processing is a relevant topic for ML problems. Environmental sounds classification is a field which attracts specialists from various spheres. Multimedia sensor networks and big variety of multimedia content, which describes urban scenes, is a huge part of it. Urban sounds classification is an important problem which needs to be solved. Many researchers focus on locations (parks, streets, etc.) instead of sources (car horn, shoot, etc.). So, urban sounds classification is a reasonable, urgent problem. Development of best-fit Machine Learning models for correct classification of urban sounds requires dealing with the following problems: sound representation as a set of numerical data considering its features for a possibility of ML algorithms to work with sound; data processing for optimization of ML models efficiency; using methods of ML models optimization for presenting the best parameters of each model; train models on training data; prediction of urban sound classes and checking the accuracy of these predictions on test data; neural networks development for solving the urban sounds classification problem.
Description
Keywords
класифікація, машинне навчання, метод k-найближчих сусідів, метод опорних векторів, випадковий ліс, нейронні мережі, контроль за k-блоками, метод відкладених даних, сітковий пошук, стаття, classification, machine learning, k-nearest neighbor, support vector machine, random forest, neural networks, k-fold cross-validation, holdout method, grid search, article
Citation
Глибовець М. М. Використання машинного навчання у задачах класифікації звуків / Глибовець М. М., Жиркова А. П. // Наукові записки НаУКМА. Комп'ютерні науки. - 2019. - Т. 2. - С. 22-31.