Speaker recognition based on characteristic spectrograms and an improved self-organizing feature map neural network

Abstract

To obtain a speaker’s pronunciation characteristics, a method is proposed based on an idea from bionics, which uses spectrogram statistics to achieve a characteristic spectrogram to give a stable representation of the speaker’s pronunciation from a linear superposition of short-time spectrograms. To deal with the issue of slow network training and recognition speed for speaker recognition systems on resource-constrained devices, based on a traditional SOM neural network, an adaptive clustering self-organizing feature map SOM (AC-SOM) algorithm is proposed. This algorithm automatically adjusts the number of neurons in the competition layer based on the number of speakers to be recognized until the number of clusters matches the number of speakers. A 100-speaker database of characteristic spectrogram samples was built and applied to the proposed AC-SOM model, yielding a maximum training time of only 304 s, with a maximum sample recognition time of less than 28 ms. Comparing to other approaches, the proposed method offers greatly improved training and recognition speed without sacrificing too much recognition accuracy. The promising results suggest that the proposed method satisfies real-time data processing and execution requirements for edge intelligence systems better than other speaker recognition methods.

Exploring the Teaching Mode of English Audiovisual Speaking in Multimedia Network Environment

“…(5) e British English pronounces |i | sound, and the American English pronounces | z |; e.g., the word system is pronounced as |'sistim| in British style and |'sistzm| in American style. (6) ere are some words that are pronounced completely differently in British English and American English. For example, leisure in British is |'leiz| and in American is |'li:zzr|.…”

“…Based on the improved deep learning convolutional network, Lin et al [5] extracted multi-scale normalized local features with stacked decreasing convolution kernel, improved the convergence speed and stability of the algorithm with dynamic learning rate, and achieved better recognition rate. Jia et al [6] used three-hidden layer deep neural network to identify MCFF features of acoustic signals and achieved better recognition results than SVM and GMM. Compared with traditional classifiers, deep learning network improves the detection accuracy of pronunciation, but its huge parameter requirements, complex parameter settings, and calculation requirements require further optimization and improvement in practical applications [7].…”

Classification of Indian Classical Music With Time-Series Matching Deep Learning Approach

“…Sa, Ri, Ga, Ma, Pa, Da, Ni are the seven musical notes carrying frequencies which further subdivided into semitones or microtones [33]. Full forms of Sa, Ri, Ga, Ma, Pa, Da, Ni are Shadja, Rishaba, Gandhara, Madhyama, Panchama, Dhaivatha and Nishadha [5].…”

Knowledge from the original network: restore a better pruned network with knowledge distillation

“…Deep neural networks have achieved excellent results that traditional machine learning can difficultly match in various fields, such as computer vision [20], point cloud processing [15,34], medical data processing [19], speech recognition [17], and so on. With the continuous development of deep learning, the artificial neural networks are becoming more and more deep, wide and complicated.…”