A Semi-supervised Speaker Identification Method for Audio Forensics Using Cochleagrams

Speech Segregation in Background Noise Based on Deep Learning

“…Many computational systems were designed to understand: speech segregation based on the ASA principles, important applications like robust speech recognition [2], and hearing aid design [10]. Pitch and amplitude modulation are examples of Computational ASA (CASA) methods applied to detached spoken rations of additive communication and the predictable tones in nearby structures were congregated using tone steadiness [3]. Cheng et al [11] engaged speaker prototypes to implement a collaborative approximation of speaker personalities and consecutive category to group temporally disconnected time-frequency (T-F) sections.…”

“…Musical subdivisions are a drawback region with acoustic substance consideration, particularly in the instance where speech isolation is essential. The precision of speech identification and isolation could be improved through noise ejections from acoustic speech indicators [3]. A learningbased approach and non-learning-based approaches were being utilized by the present techniques of speech and music isolation [2].…”

A hybrid technique for speech segregation and classification using a sophisticated deep neural network

“…Noise garbles speech and introduces obstacles in various applications, including automatic speech segregation. Noise removal from audio speech signals enhance the accuracy of speech recognition and segregation applications [ 2 ].…”

“…Learning-based methods are employed more frequently than non-learning-based methods because of their potential for segregating speech and music components more effectively in the presence of background noise. Lekshmil and Sathidevi [ 1 ] proposed non-learning-based speech segregation models for single-channel speech separation using short-time Fourier transform (STFT) [ 2 ]. They use pitch information-based techniques for the segregation process.…”

“…They use pitch information-based techniques for the segregation process. Time-frequency mask-based pitch frequencies are gathered using dominant and interfering speaker information [ 2 ]. Cheng et al and Hu et al [ 3 , 4 ] proposed co-channel speech segregation using a non-learning-based model.…”