Sparse Representation with Optimized Learned Dictionary for Robust Voice Activity Detection

Enhanced Speech Based Jointly Statistical Probability Distribution Function for Voice Activity Detection

“…After that, we also found that the mutual coherence between the two learned dictionary is an important adverse factor for robustness enhancing [23] . For enhancing the robustness of concatenated dictionary based sparse representation, our previous work has provided an optimization algorithm to reduce the mutual coherence between the learned speech dictionary and noise dictionary, and its experimental results showed that the optimization algorithm can significantly reduces the mutual coherence between the concatenated dictionaries and improves the robustness of the sparse representation [23] . For going further to improve the robustness of VAD under non-stationary noise conditions, we propose a robust VAD method based on the previous work in this paper.…”

“…Owing to the atoms of learned speech dictionary, which are well adaptable to speech underlying structures, observed speech signal can be effectively decomposed by sparse representation in the learned dictionary; similarly, the learned noise dictionary is more adaptable to the noise underlying structures, and provides good basis for sparse representation of noise. It is known that the underlying structures of speech signal and noise signal are different, we found that when a learned speech dictionary and a learned noise dictionary are concatenated as an over-completed dictionary for sparse representation, then the robustness outperforms single dictionary based sparse representation [23] . After that, we also found that the mutual coherence between the two learned dictionary is an important adverse factor for robustness enhancing [23] .…”

“…It is known that the underlying structures of speech signal and noise signal are different, we found that when a learned speech dictionary and a learned noise dictionary are concatenated as an over-completed dictionary for sparse representation, then the robustness outperforms single dictionary based sparse representation [23] . After that, we also found that the mutual coherence between the two learned dictionary is an important adverse factor for robustness enhancing [23] . For enhancing the robustness of concatenated dictionary based sparse representation, our previous work has provided an optimization algorithm to reduce the mutual coherence between the learned speech dictionary and noise dictionary, and its experimental results showed that the optimization algorithm can significantly reduces the mutual coherence between the concatenated dictionaries and improves the robustness of the sparse representation [23] .…”

A Voice Activity Detection Algorithm Using Sparse Non-negative Matrix Factorization-based Model Learning in Spectro-Temporal Domain

“…The labels of the different parts of the input frame are determined by calculating the energy in these frames. You et al [4] optimized algorithm for learning speech and noise dictionaries. The goal of this optimization procedure was to reduce the coherence value between the learned dictionaries to obtain a robust VAD algorithm in the different noise conditions.…”