Gammatone spectral latitude features extraction for pathological voice detection and classification

Acoustic recognition was always used in fault diagnosis in mechanical equipment and observation systems. For the operated transformer, acoustic signals were mainly originated from mechanical vibrations of winding and core and then emitted in the air. Consequently, abundant information about mechanical structures and running conditions of transformer was contained in acoustic signals. Based on the better recognition ability and robust of human hearing system, cochlear filter cepstral coefficients (CFCCs) of acoustic signals of transformer was calculated in the paper. Through the construction of training set and testing set from the CFCCs, the probabilistic neural network (PNN) was trained to recognize the typical mechanical faults of transformers. The load experiment of a dry type transformer with the voltage rating of 10kV was made in different load power when the transformer is in normal and typical mechanical faults conditions. Calculated results have shown that the proposed method is capable of describing the features of acoustic signals and recognizing the mechanical faults of power transformer with high accuracy.

Cochlear Filter Cepstral Coefficients of Acoustic Signals for Mechanical Faults Identification of Power Transformer

Isolated word recognition based on PNCC with different classifiers in a noisy environment

The end-to-end speech recognition technology solves the problem that each component is independent and models cannot be jointly optimized in the traditional speech recognition model. It incorporates such components as the acoustic model, language model, and decoding unit of the hybrid model into a single neural network, that can avoid the inherent defects of multiple modules and greatly reduces the complexity of the speech recognition model. In this research, an Amdo-Tibetan speech recognition system is constructed based on Listen, Attend and Spell (LAS) model by the end-to-end speech recognition technology. It can realize the direct conversion from Amdo-Tibetan speech sequence to the corresponding character sequence and greatly reduces the difficulty of building the Amdo-Tibetan speech recognition model. To further improve the performance of the proposed system, the following improvements have been made: firstly, the Multi-Head Attention mechanism is introduced to improve the alignment accuracy between state vectors of decoder and encoder; secondly, the label smoothing technique is adopted to solve the problem of over-fitting; thirdly, an N-gram language model is combined with the LAS model to increase the accuracy of speech recognition and the maximum mutual information (MMI) criterion is employed for discriminative training; and finally, transfer learning is utilized to overcome the problem of insufficient training data. Experimental results show that the proposed model can significantly enhance the performance of Amdo-Tibetan speech recognition.

/pdf/end-to-end-amdo-tibetan-speech-recognition-based-on-tp9rcwvo28.pdf

End-to-End Amdo-Tibetan Speech Recognition Based on Knowledge Transfer

Environmental noise can pose a threat to the stable operation of current speech recognition systems. It is therefore essential to develop a front feature set that is able to identify speech under low signal-to-noise ratio. In this paper, a robust fusion feature is proposed that can fully characterize speech information. To obtain the cochlear filter cepstral coefficients (CFCC), a novel feature is first extracted by the power-law nonlinear function, which can simulate the auditory characteristics of the human ear. Speech enhancement technology is then introduced into the front end of feature extraction, and the extracted feature and their first-order difference are combined in new mixed features. An energy feature Teager energy operator cepstral coefficient (TEOCC) is also extracted, and combined with the above-mentioned mixed features to form the fusion feature sets. Principal component analysis (PCA) is then applied to feature selection and optimization of the feature set, and the final feature set is used in a non-specific persons, isolated words, and small-vocabulary speech recognition system. Finally, a comparative experiment of speech recognition is designed to verify the advantages of the proposed feature set using a support vector machine (SVM). The experimental results show that the proposed feature set not only display a high recognition rate and excellent anti-noise performance in speech recognition, but can also fully characterize the auditory and energy information in the speech signals.

/pdf/fusion-feature-extraction-based-on-auditory-and-energy-for-3o3sgr1hkm.pdf

Fusion Feature Extraction Based on Auditory and Energy for Noise-Robust Speech Recognition

The invention belongs to the speech recognition technological field and relates to an anti-noise speech recognition system. The implementation process of the system includes the following steps that:windowing and framing are performed on speech signals, discrete Fourier transformation is performed, the amplitude and phase angle of the speech signals can be obtained; the power spectrum of estimated signals is obtained by means of spectral subtraction operation; the signals are reconstructed through using pre-spectral subtraction phase angle information, so that a post-spectral subtraction speech sequence can be obtained; a non-linear power function is adopted to simulate human ear auditory features on the basis of the new speech sequence so as to extract a cochlear filter cepstral coefficient and the first-order difference deltaCFCC thereof, and feature fusion is carried out through using a dimension screening method; data normalization processing is performed on fused features, so that training set labels and test set labels are obtained; and PCA is adopted to perform dimensionality reduction on the normalized training set, and the dimensionality-reduced training set is substituted into an SVM model, so that recognition accuracy can be obtained.

Anti-noise speech recognition system

The invention relates to the speech recognition technology field, and discloses an adaptive mutation bird swarm speech recognition system integrated with data normalization. A bird swarm individual updating method is optimized by adopting an adaptive mutation method. In the iterative updating initial stage of the bird swarm algorithm, mutation operation is introduced to optimize a parameter adaptive process, and by combining with the data normalization method, a data structure and algorithm complexity are simplified, the swarm diversity of the algorithm is effectively improved, a generalization capability of a model is enhanced, the problems of the algorithm of premature convergence and gradual decreasing of a searching capability along with increasing of generation numbers are solved, and a defect of inclination to local optimization is improved. The adaptive mutation bird swarm speech recognition system has advantages of higher recognition accuracy, faster convergence speed, stronger robustness, and better optimization searching effect.

Xue Peiyun

Papers

Fusion Feature Extraction Based on Auditory and Energy for Noise-Robust Speech Recognition

Anti-noise speech recognition system

Adaptive mutation bird swarm speech recognition system integrated with data normalization