Unsupervised voice activity detection with improved signal-to-noise ratio in noisy environment. (24th May 2023)

• Record Type: Journal Article
• Authors: Sharma, Shilpa; Malhotra, Rahul; Sharma, Anurag; Bala, Jeevan; Rattan, Punam; Vashisht, Sheveta
• Is Part Of: International journal of nanotechnology. Volume 20:Number 1/4(2023)
• Page Start: 421
• Page End: 432
• Publication Date: 2023-05-24
• Subjects: TIMIT dataset -- support vector machine -- voice activity detector -- unsupervised learning; 620.505
• Languages: English
• ISSNs: 1475-7435

Abstract

To identify voiced and unvoiced signals, this research provides an extended voice characteristic detection strategy for noisy settings that uses feature extraction and unvoiced feature normalisation. In a high signal to noise ratio environment, the proposed method develops a recognition model by recovering characteristics for categorisation of spoken and unvoiced signals. The novelty of this method is that it uses feature extraction to classify voiced and unvoiced signals with a higher signal-to-noise ratio (SNR). Furthermore, by combining two classifiers in a hybrid model, the model is less affected by noise for speech features, and identification performance improves. The model was tested for its ability to increase recognition accuracy. The proposed method produces better results than existing methods, with an accuracy of 99.73% and SNR of 25.61 dB. The proposed model LFV-KANN also handles increases in noise power efficiently through the hybridisation of two classifiers: artificial neural network (ANN) and K-means clustering.