To search, Click below search items. |
| All Published Papers Search Service |
|
|
|
Title |
A Comparison of Machine Learning Methods using Correlated Speech Features in the Presence of Varied Noise |
|
Author |
D. U. R Khan, Syed Abbas Ali, and Hina D. Khan |
| Citation |
Vol. 22 No. 4 pp. 535-544 |
|
Abstract |
Speech signal analysis processing helps extract information from both clean and noisy speech signals, and machine learning algorithms provide robust analytical tools for signal exploration. In this research (14) speech signal features were analyzed using machine learning tools with the following corpuses of speech commands: clean speech, with average noise, and with high noise. The analysis is based on the selection of the most correlated feature of distant and noisy speech along with the implementation of three (03) conventional learning (random forest nearest neighbor, voting model, and support vector machine (SVM)) and deep learning (Long short-term memory (LSTM)) models. This study presents a comprehensive result of selected features with clean, average noise, and very high-noise speech corpuses. The respective signal features performed well with a support vector machine (SVM) with no noise and average noise corpuses. However, LSTM shows significant results with high-noise corpus inters with macro-and average-weighted accuracy. |
|
Keywords |
learning algorithms, LSTM, robust speech, speech |
|
URL |
