Browsing by Author "Shen, Yih-Liang"

Now showing 1 - 3 of 3

A GENERATIVE AUDITORY MODEL EMBEDDED NEURAL NETWORK FOR SPEECH PROCESSING
(2018-01-01)
Lo, Yu-Wen
;
Shen, Yih-Liang
;
Liao, Yuan-Fu
;
Chi, Tai-Shih
;
電機工程學系
Before the era of the neural network (NN), features extracted from auditory models have been applied to various speech applications and been demonstrated more robust against noise than conventional speech-processing features. What's the role of auditory models in the current NN era? Are they obsolete? To answer this question, we construct a NN with a generative auditory model embedded to process speech signals. The generative auditory model consists of two stages, the stage of spectrum estimation in the logarithmic-frequency axis by the cochlea and the stage of spectral-temporal analysis in the modulation domain by the auditory cortex. The NN is evaluated in a simple speaker identification task. Experiment results show that the auditory model embedded NN is still more robust against noise, especially in low SNR conditions, than the randomly-initialized NN in speaker identification.
Plastic multi-resolution auditory model based neural network for speech enhancement
(2017-01-01)
Lai, Chen-Yen
;
Lo, Yu-Wen
;
Shen, Yih-Liang
;
Chi, Tai-Shih
;
電機工程學系
In this paper, we propose a plastic auditory model based neural network for speech enhancement. The proposed system integrates a spectro-temporal analytical auditory model with a multi-layer fully-connected network to form a quasi-CNN structure. The initial kernels of the convolutional layer are derived from the neuro-physiological auditory model. To simulate the plasticity of cortical neurons for attentional hearing, the kernels are allowed to adjust themselves pertaining to the task at hand. For the application of speech enhancement, the Fourier spectrogram instead of the auditory spectrogram is used as input to the proposed neural network such that the cleaned speech signal can be well reconstructed. The proposed system performs comparably with standard DNN and CNN systems when plenty resources are available. Meanwhile, under the limited-resource condition, the proposed system outperforms standard systems in all test settings.
REINFORCEMENT LEARNING BASED SPEECH ENHANCEMENT FOR ROBUST SPEECH RECOGNITION
(2019-01-01)
Shen, Yih-Liang
;
Huang, Chao-Yuan
;
Wang, Syu-Siang
;
Tsao, Yu
;
Wang, Hsin-Min
Conventional deep neural network (DNN)-based speech enhancement (SE) approaches aim to minimize the mean square error (MSE) between enhanced speech and clean reference. The MSE-optimized model may not directly improve the performance of an automatic speech recognition (ASR) system. If the target is to minimize the recognition error, the recognition results should be used to design the objective function for optimizing the SE model. However, the structure of an ASR system, which consists of multiple units, such as acoustic and language models, is usually complex and not differentiable. In this study, we propose to adopt the reinforcement learning (RL) algorithm to optimize the SE model based on the recognition results. We evaluated the proposed RL-based SE system on the Mandarin Chinese broadcast news corpus (MATBN). Experimental results demonstrate that the proposed SE system can effectively improve the ASR results with a notable 12 : 40% and 19 : 23% error rate reductions for signal to noise ratio (SNR) at 0 dB and 5 dB conditions, respectively.