This paper proposes a deep learning-based non-intrusive objective speech intelligibility estimation method based on recurrent neural network (RNN) with long short-term memory (LSTM) structure. Conventional non-intrusive estimation methods such as standard P.563 have poor estimation performance and lack of consistency, especially, in various noise and reverberation environments. The proposed method trains the LSTM RNN model parameters by utilizing the STOI that is the standard intrusive intelligibility estimation method with reference speech signal. The input and output of the LSTM RNN are the MFCC vector and the frame-wise STOI value, respectively. Experimental results show that the proposed objective intelligibility estimation method outperforms the conventional standard P.563 in various noisy and reverberant environments.
This paper proposes an audio data augmentation method based on deep learning in order to improve the performance of dereverberation. Conventionally, audio data are augmented using a room impulse response, which is artificially generated by some methods, such as the image method. The proposed method estimates a reverberation environment model based on a deep neural network that is trained by using clean and recorded audio data as inputs and outputs, respectively. Then, a large amount of a real augmented database is constructed by using the trained reverberation model, and the dereverberation model is trained with the augmented database. The performance of the augmentation model was verified by a log spectral distance and mean square error between the real augmented data and the recorded data. In addition, according to dereverberation experiments, the proposed method showed improved performance compared with the conventional method.
This paper presents a deep learning-based non-intrusive speech intelligibility estimation method using bottleneck features of autoencoder. The conventional standard non-intrusive speech intelligibility estimation method, P.563, lacks intelligibility estimation performance in various noise environments. We propose a more accurate speech intelligibility estimation method based on long-short term memory (LSTM) neural network whose input and output are an autoencoder bottleneck features and a short-time objective intelligence (STOI) score, respectively, where STOI is a standard tool for measuring intrusive speech intelligibility with reference speech signals. We showed that the proposed method has a superior performance by comparing with the conventional standard P.563 and mel-frequency cepstral coefficient (MFCC) feature-based intelligibility estimation methods for speech signals in various noise environments.
No abstract
It is well known that one of the major advantage of neural networks lies in the constant time response once they are trained. However, if a neural network includes an MAXNET [5], a neural net for selection of maximum values, it usually cannot get the constant time response due to its drawback of inconsistent convergent time. Unfortunately, a neural network for classification often needs to use an MAXNET as a subnet generally.As a result, MAXNET can be a bottleneck in the recalling phase so that it is impractical to combine it with other neural networks. Most importantly, when MAXNET serves as an intermechary subnet. the neural network may be led to a wrong decision in the case that there are several largest values in the intermediate results, due to the random selection among those values of MAXNET. In this paper, an MAXNET with constant-time response and with miiltiple selection of several maximum values, CTMAXNET. is presented. The constant-time response for CTMAXNET is proved. The cost analysis of VLSI implementation for CTMAXNET is presented. It is compared with the cost of MAXNET in [6]. The result shows that CTMAXNET has less cost than MAX-NET. From the example of ARTI, it shows that including MAXNET leads to misclassification while including CTMAXNET does not. Through these results, a case is made that CTMAXNET is worthy of imbedding in a large number of neural network classifier models and applications.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.