基于深度神经网络的方法已经在语声增强领域得到了广泛的应用,然而若想取得理想的性能,一般需要规模较大且复杂度较高的模型。因此,在计算资源有限的设备或对延时要求高的环境下容易出现部署困难的问题。为了解决此问题,提出了一种基于...基于深度神经网络的方法已经在语声增强领域得到了广泛的应用,然而若想取得理想的性能,一般需要规模较大且复杂度较高的模型。因此,在计算资源有限的设备或对延时要求高的环境下容易出现部署困难的问题。为了解决此问题,提出了一种基于深度复卷积递归网络的师生学习语声增强方法。在师生深度复卷积递归网络模型结构中间的复长短时记忆递归模块提取实部和虚部特征流,并分别计算帧级师生距离损失以进行知识转移。同时使用多分辨率频谱损失以进一步提升低复杂度学生模型的性能。实验在公开数据集Voice Bank Demand和DNS Challenge上进行,结果显示所提方法相对于基线学生模型在各项指标上均有明显提升。展开更多
An improved speech absence probability estimation was proposed using environmental noise classification for speech enhancement.A relevant noise estimation approach,known as the speech presence uncertainty tracking met...An improved speech absence probability estimation was proposed using environmental noise classification for speech enhancement.A relevant noise estimation approach,known as the speech presence uncertainty tracking method,requires seeking the "a priori" probability of speech absence that is derived by applying microphone input signal and the noise signal based on the estimated value of the "a posteriori" signal-to-noise ratio(SNR).To overcome this problem,first,the optimal values in terms of the perceived speech quality of a variety of noise types are derived.Second,the estimated optimal values are assigned according to the determined noise type which is classified by a real-time noise classification algorithm based on the Gaussian mixture model(GMM).The proposed algorithm estimates the speech absence probability using a noise classification algorithm which is based on GMM to apply the optimal parameter of each noise type,unlike the conventional approach which uses a fixed threshold and smoothing parameter.The performance of the proposed method was evaluated by objective tests,such as the perceptual evaluation of speech quality(PESQ) and composite measure.Performance was then evaluated by a subjective test,namely,mean opinion scores(MOS) under various noise environments.The proposed method show better results than existing methods.展开更多
A filter algorithm based on cochlear mechanics and neuron filter mechanism is proposed from the view point of vibration.It helps to solve the problem that the non-linear amplification is rarely considered in studying ...A filter algorithm based on cochlear mechanics and neuron filter mechanism is proposed from the view point of vibration.It helps to solve the problem that the non-linear amplification is rarely considered in studying the auditory filters.A cochlear mechanical transduction model is built to illustrate the audio signals processing procedure in cochlea,and then the neuron filter mechanism is modeled to indirectly obtain the outputs with the cochlear properties of frequency tuning and non-linear amplification.The mathematic description of the proposed algorithm is derived by the two models.The parameter space,the parameter selection rules and the error correction of the proposed algorithm are discussed.The unit impulse responses in the time domain and the frequency domain are simulated and compared to probe into the characteristics of the proposed algorithm.Then a 24-channel filter bank is built based on the proposed algorithm and applied to the enhancements of the audio signals.The experiments and comparisons verify that,the proposed algorithm can effectively divide the audio signals into different frequencies,significantly enhance the high frequency parts,and provide positive impacts on the performance of speech enhancement in different noise environments,especially for the babble noise and the volvo noise.展开更多
文摘基于深度神经网络的方法已经在语声增强领域得到了广泛的应用,然而若想取得理想的性能,一般需要规模较大且复杂度较高的模型。因此,在计算资源有限的设备或对延时要求高的环境下容易出现部署困难的问题。为了解决此问题,提出了一种基于深度复卷积递归网络的师生学习语声增强方法。在师生深度复卷积递归网络模型结构中间的复长短时记忆递归模块提取实部和虚部特征流,并分别计算帧级师生距离损失以进行知识转移。同时使用多分辨率频谱损失以进一步提升低复杂度学生模型的性能。实验在公开数据集Voice Bank Demand和DNS Challenge上进行,结果显示所提方法相对于基线学生模型在各项指标上均有明显提升。
基金Project supported by an Inha University Research GrantProject(10031764) supported by the Strategic Technology Development Program of Ministry of Knowledge Economy,Korea
文摘An improved speech absence probability estimation was proposed using environmental noise classification for speech enhancement.A relevant noise estimation approach,known as the speech presence uncertainty tracking method,requires seeking the "a priori" probability of speech absence that is derived by applying microphone input signal and the noise signal based on the estimated value of the "a posteriori" signal-to-noise ratio(SNR).To overcome this problem,first,the optimal values in terms of the perceived speech quality of a variety of noise types are derived.Second,the estimated optimal values are assigned according to the determined noise type which is classified by a real-time noise classification algorithm based on the Gaussian mixture model(GMM).The proposed algorithm estimates the speech absence probability using a noise classification algorithm which is based on GMM to apply the optimal parameter of each noise type,unlike the conventional approach which uses a fixed threshold and smoothing parameter.The performance of the proposed method was evaluated by objective tests,such as the perceptual evaluation of speech quality(PESQ) and composite measure.Performance was then evaluated by a subjective test,namely,mean opinion scores(MOS) under various noise environments.The proposed method show better results than existing methods.
基金Project(17KJB510029)supported by the Natural Science Foundation of the Jiangsu Higher Education Institutions,ChinaProject(GXL2017004)supported by the Scientific Research Foundation of Nanjing Forestry University,China+3 种基金Project(202102210132)supported by the Important Project of Science and Technology of Henan Province,ChinaProject(B2019-51)supported by the Scientific Research Foundation of Henan Polytechnic University,ChinaProject(51521003)supported by the Foundation for Innovative Research Groups of the National Natural Science Foundation of ChinaProject(KQTD2016112515134654)supported by Shenzhen Science and Technology Program,China。
文摘A filter algorithm based on cochlear mechanics and neuron filter mechanism is proposed from the view point of vibration.It helps to solve the problem that the non-linear amplification is rarely considered in studying the auditory filters.A cochlear mechanical transduction model is built to illustrate the audio signals processing procedure in cochlea,and then the neuron filter mechanism is modeled to indirectly obtain the outputs with the cochlear properties of frequency tuning and non-linear amplification.The mathematic description of the proposed algorithm is derived by the two models.The parameter space,the parameter selection rules and the error correction of the proposed algorithm are discussed.The unit impulse responses in the time domain and the frequency domain are simulated and compared to probe into the characteristics of the proposed algorithm.Then a 24-channel filter bank is built based on the proposed algorithm and applied to the enhancements of the audio signals.The experiments and comparisons verify that,the proposed algorithm can effectively divide the audio signals into different frequencies,significantly enhance the high frequency parts,and provide positive impacts on the performance of speech enhancement in different noise environments,especially for the babble noise and the volvo noise.
