Study on the Use and Adaptation of Bottleneck Features for Robust Speech Recognition of Nonlinearly Distorted Speech
Jiří Málek, Petr Červa, Ladislav Šeps, Jan Nouza · 2016
This paper focuses on the robust recognition of nonlinearly distorted speech. We have reported (Seps et al., 2014) that hybrid acoustic models based on a combination of Hidden Markov Models and Deep Neural Networks (HMM-DNNs) are better suited to this task than conventional HMMs utilizing Gaussian Mixture Models (HMM-GMMs). To further improve recognition accuracy, this paper investigates the possibility of combining the modeling power of deep neural networks with the adaptation to given acoustic conditions. For this purpose, the deep neural networks are utilized to produce bottleneck coefficients / features (BNC). The BNCs are subsequently used for training of HMM-GMM based acoustic models and then adapted using Constrained Maximum Likelihood Linear Regression (CMLLR). Our results obtained for three types of nonlinear distortions and three types of input features show that the adapted BNC-based system (a) outperforms HMM-DNN acoustic models in the case of strong compression and (b) yields comparable performance for speech affected by nonlinear amplification in the analog domain.