Distant-talking accent recognition by combining GMM and DNN

Abstract

Recently, automatic accent recognition has been paid more and more attentions. However, there are few researches focusing on accent recognition in distant-talking environment which is very important for improving distant-talking speech recognition performance with non-native accents. In this paper, we apply Gaussian Mixture Models (GMM) and Deep Neural Network (DNN) to identify the speaker accent in reverberant environments. The combination of likelihood with these two approaches is also proposed. In reverberant environment, the accent recognition rate was improved from 90.7 % with GMM to 93.0 % with DNN. The combination of GMM and DNN achieved recognition rate of 97.5 %, which outperformed than the individual GMM and DNN because the complementation of GMM and DNN. The relative error reduction is 73.1 % than the GMM-based method and 64.3 % than the DNN-based method, respectively.

Spectrogram based multi-task audio classification

“…Additionally, LSTM was used in a hybrid emotion inference model that was proposed for inferring user emotion in a real-world voice-dialogue application, and a recurrent autoencoder was proposed to pre-train the LSTM to improve accuracy [32]. Further, GMM and DNNs were combined to identify distant accents in reverberant environments [26]. The authors found that this combination of classifiers outperformed the individual GMM and DNNs classifiers.…”

Exploiting Magnitude and Phase Aware Deep Neural Network for Replay Attack Detection

“…Although MP-aware DNN-based detection may provide a better performance than that of conventional DNN-based detection using only magnitude information, it may not work well due to the lack of feature-based resolution within low frequencies [17] and limited training data. In [18], a combination of GMM and DNN for distant-talking accent recognition was proposed to increase the accent recognition performance on limited training data. The result showed that the combination of these two different methods could improve the distant-accent recognition performance when compared to just an individual one.…”

“…The result showed that the combination of these two different methods could improve the distant-accent recognition performance when compared to just an individual one. Motivated by [18], a combination of CQCC-based GMM and MP-aware DNN was proposed to take advantage of these benefits of different classifications and features and was expected to achieve better performance. The probabilities obtained from the different systems are combined by the following equation.…”

“…The applied MP-aware DNN-based detection was compared with conventional DNN-based detection and baseline constant Q transform cepstral coefficients-based Gaussian mixture classification (CQCC-based GMM). Although the MP-aware DNN-based method may provide better discrimination detection than conventional DNN-based detection using only magnitude/phase information, it may not work well due to the lack of feature-based reso-lution within low frequencies [17] and limited training data [18]. To address this problem, a novel method was proposed; combining MP-aware DNN with CQCC-based GMM to improve the reliable detection decision.…”

Guest Editorial: Immersive Audio/Visual Systems

“…In this special issue, Ren et al [10] propose three integration schemes for robust distant-talking speech recognition which combine bottleneck feature extraction with dereverberation technique. As an accompanying paper by the same institution, Phapatanaburi et al [9] propose a combination of Gaussian Mixture Models (GMM) and Deep Neural Networks (DNNs) to identify the speaker accent in reverberant environments.…”