Voice Activity Detection for Transient Noisy Environment Based on Diffusion Nets

Abstract

We address voice activity detection in acoustic environments of transients and stationary noises, which often occur in real life scenarios. We exploit unique spatial patterns of speech and non-speech audio frames by independently learning their underlying geometric structure. This process is done through a deep encoder-decoder based neural network architecture. This structure involves an encoder that maps spectral features with temporal information to their low-dimensional representations, which are generated by applying the diffusion maps method. The encoder feeds a decoder that maps the embedded data back into the high-dimensional space. A deep neural network, which is trained to separate speech from non-speech frames, is obtained by concatenating the decoder to the encoder, resembling the known Diffusion nets architecture. Experimental results show enhanced performance compared to competing voice activity detection methods. The improvement is achieved in both accuracy, robustness and generalization ability. Our model performs in a real-time manner and can be integrated into audio-based communication systems. We also present a batch algorithm which obtains an even higher accuracy for off-line applications.

Evaluation of Deep-Learning-Based Voice Activity Detectors and Room Impulse Response Models in Reverberant Environments

“…An interesting outcome shows in each of the tested setups, the leading accuracy of each detector was consistently achieved by the Valeau RIR model [18]. In a similar manner, the detector introduced in Ivry [12] prevailed competing VADs across all experiments.…”

“…This study has also shown that the response model introduced by Valeau [18] consistently leads to the best performance, regardless of the detector and the tested acoustic environment. That and more, the VAD introduced by Ivry [12] has achieved leading performance across all experiments. In future work, additional aspects such as feature engineering and dedicated architecture will be addressed in order to further enhance Ivry detector and adjust it for practical and reverberant acoustic scenarios.…”

“…In this study, we consider five VADs [8][9][10][11][12] and address them as Ariav-R, Wagner, Kim, Ariav-W and Ivry, respectively. Also, we employ five RIR models [14][15][16][17][18], and refer them as Borish, Vorländer, Randel, Lam and Valeau, correspondingly.…”

“…Next, we focus on Ivry VAD [12] that achieved leading performance in all previous experiments, as can be viewed in Figs. 1 and 2.…”

“…Combined end-to-end VAD system was introduced by Ariav et al [11], that comprised of WaveNet for feature extraction and a deep residual network for speech detection. Ivry et al [12] applied ensemble learning with two deep encoder-decoder structures to learn the unique temporal and spatial patterns of speech through the diffusion maps method.…”