Search

Fake audio detection (FAD) is a technique to distinguish synthetic speech from natural speech. In most FAD systems, removing irrelevant features from acoustic speech while keeping only robust discriminative features is essential. Intuitively, speaker information entangled in acoustic speech should be suppressed for the FAD task. Particularly in a deep neural network (DNN)-based FAD system, the learning system may learn speaker information from a training dataset and cannot generalize well on a testing dataset. In this paper, we propose to use the speaker anonymization (SA) technique to suppress speaker information from acoustic speech before inputting it into a DNN-based FAD system. We adopted the McAdamscoefficient-based SA (MC-SA) algorithm, and this is expected that the entangled speaker information will not be involved in the DNN-based FAD learning. Based on this idea, we implemented a light convolutional neural network bidirectional long short-term memory (LCNN-BLSTM)-based FAD system and conducted experiments on the Audio Deep Synthesis Detection Challenge (ADD2022) datasets. The results showed that removing the speaker information from acoustic speech improved the relative performance in the first track of ADD2022 by 17.66%., Interspeech 2022, 18-22 September 2022, Incheon, Korea, identifier:https://dspace.jaist.ac.jp/dspace/handle/10119/18158

Fake audio detection (FAD) is a technique to distinguish synthetic speech from natural speech. In most FAD systems, removing irrelevant features from acoustic speech while keeping only robust discriminative features is essential. Intuitively, speaker information entangled in acoustic speech should be suppressed for the FAD task. Particularly in a deep neural network (DNN)-based FAD system, the learning system may learn speaker information from a training dataset and cannot generalize well on a testing dataset. In this paper, we propose to use the speaker anonymization (SA) technique to suppress speaker information from acoustic speech before inputting it into a DNN-based FAD system. We adopted the McAdamscoefficient-based SA (MC-SA) algorithm, and this is expected that the entangled speaker information will not be involved in the DNN-based FAD learning. Based on this idea, we implemented a light convolutional neural network bidirectional long short-term memory (LCNN-BLSTM)-based FAD system and conducted experiments on the Audio Deep Synthesis Detection Challenge (ADD2022) datasets. The results showed that removing the speaker information from acoustic speech improved the relative performance in the first track of ADD2022 by 17.66%., Interspeech 2022, 18-22 September 2022, Incheon, Korea, identifier:https://dspace.jaist.ac.jp/dspace/handle/10119/18158

Fake audio detection (FAD) is a technique to distinguish synthetic speech from natural speech. In most FAD systems, removing irrelevant features from acoustic speech while keeping only robust discriminative features is essential. Intuitively, speaker information entangled in acoustic speech should be suppressed for the FAD task. Particularly in a deep neural network (DNN)-based FAD system, the learning system may learn speaker information from a training dataset and cannot generalize well on a testing dataset. In this paper, we propose to use the speaker anonymization (SA) technique to suppress speaker information from acoustic speech before inputting it into a DNN-based FAD system. We adopted the McAdamscoefficient-based SA (MC-SA) algorithm, and this is expected that the entangled speaker information will not be involved in the DNN-based FAD learning. Based on this idea, we implemented a light convolutional neural network bidirectional long short-term memory (LCNN-BLSTM)-based FAD system and conducted experiments on the Audio Deep Synthesis Detection Challenge (ADD2022) datasets. The results showed that removing the speaker information from acoustic speech improved the relative performance in the first track of ADD2022 by 17.66%., Interspeech 2022, 18-22 September 2022, Incheon, Korea

Voice activity detection (VAD) is used to detect whether the acoustic signal belongs to speech or non-speech clusters based on the statistical distribution of the acoustic features. Traditional VAD algorithms are applied in a linear transformed space without any constraint relating to the special characteristics speech or noise. As a result, the VAD algorithms are not robust to noise interference. Considering that speech is a special type of acoustic signal that only occupies a small fraction of the whole acoustic space, we proposed a new speech feature extraction method by giving constraints on the processing space as a reproducing kernel Hilbert space (RKHS). In the RKHS, we regarded the speech estimation as a functional approximation problem, and estimated the approximation function via a regularized framework in the RKHS. Under this framework, we could incorporate the nonlinear mapping functions in the approximation implicitly via a kernel function. The approximation function could capture the nonlinear and high-order statistical regularities of the speech. Our VAD algorithm is designed on the basis of the power energy in this regularized RKHS. Compared with a baseline and G.729B VAD algorithms, experimental results showed the promising advantages of our proposed algorithm., identifier:https://dspace.jaist.ac.jp/dspace/handle/10119/9580

Voice activity detection (VAD) is used to detect whether the acoustic signal belongs to speech or non-speech clusters based on the statistical distribution of the acoustic features. Traditional VAD algorithms are applied in a linear transformed space without any constraint relating to the special characteristics speech or noise. As a result, the VAD algorithms are not robust to noise interference. Considering that speech is a special type of acoustic signal that only occupies a small fraction of the whole acoustic space, we proposed a new speech feature extraction method by giving constraints on the processing space as a reproducing kernel Hilbert space (RKHS). In the RKHS, we regarded the speech estimation as a functional approximation problem, and estimated the approximation function via a regularized framework in the RKHS. Under this framework, we could incorporate the nonlinear mapping functions in the approximation implicitly via a kernel function. The approximation function could capture the nonlinear and high-order statistical regularities of the speech. Our VAD algorithm is designed on the basis of the power energy in this regularized RKHS. Compared with a baseline and G.729B VAD algorithms, experimental results showed the promising advantages of our proposed algorithm.

This paper proposes two methods for robust automatic speech recognition (ASR) in reverberant environments. Unlike other methods which mostly apply inverse filtering by blindly estimated room impulse responses to achieve dereverberation, theproposed methods are based on the utilization of the characteristics of speech. The first method - Harmonicity based Feature Analysis – takes advantage of the harmonic componentsof speech, which are assumed to be undistorted. The second method - Temporal Power Envelope Feature Analysis – utilizes the temporal modulation structure of speech, representing the phoneme level temporal events which contain most intelligibility information. Both methods increase the recognition performance remarkably in a different way. Combining both of them connects their individual advantages. In order to examine theperformance of utilizing harmonicity and modulation temporal structure for reverberant ASR, the methods are tested in clean and reverberant training. As results show, even in strong reverberantconditions both methods obtain practical applicableperformance for reverberant training. In addition, besides testing their performance in dependency on the reverberation time, their performance considering the speaker-to-microphone distanceis tested, which is another new contributions in this paper.