Discriminative Learning of Filterbank Layer within Deep Neural Network Based Speech Recognition for Speaker Adaptation

Abstract

Deep neural networks (DNNs) have achieved significant success in the field of automatic speech recognition. One main advantage of DNNs is automatic feature extraction without human intervention. However, adaptation under limited available data remains a major challenge for DNN-based systems because of their enormous free parameters. In this paper, we propose a filterbank-incorporated DNN that incorporates a filterbank layer that presents the filter shape/center frequency and a DNN-based acoustic model. The filterbank layer and the following networks of the proposed model are trained jointly by exploiting the advantages of the hierarchical feature extraction, while most systems use predefined mel-scale filterbank features as input acoustic features to DNNs. Filters in the filterbank layer are parameterized to represent speaker characteristics while minimizing a number of parameters. The optimization of one type of parameters corresponds to the Vocal Tract Length Normalization (VTLN), and another type corresponds to feature-space Maximum Linear Likelihood Regression (fMLLR) and feature-space Discriminative Linear Regression (fDLR). Since the filterbank layer consists of just a few parameters, it is advantageous in adaptation under limited available data. In the experiment, filterbank-incorporated DNNs showed effectiveness in speaker/gender adaptations under limited adaptation data. Experimental results on CSJ task demonstrate that the adaptation of proposed model showed 5.8% word error reduction ratio with 10 utterances against the un-adapted model.

Adaptive Multi-layer Distributed System Based on Neural Network

“…Multi-layer distributed system is developed from the original one-layer to two-layer distributed system to the current three-layer or even more layer distributed system. A new layer is added to the two-layer application service layer, thus forming a multi-layer distributed system, which is mainly composed of interface layer, transaction layer and data access layer [4]. The specific functions are as follows,as shown in Figure 1:…”

Satellite Cluster Fault Diagnosis and Detection Analysis Based on Neural Network

“…In order to verify the effectiveness of the neural network for fault diagnosis and detection of satellite swarms, the simulation constellation parameters shown in the following table are used in the experiment [14]. The FEO layer is composed of 6 high-orbit satellites (GEO) the orbits are connected by LOU, and the LEL is composed of polar orbit constellations [15]. As shown in Table 1…”

Low-Level Physiological Implications of End-to-End Learning for Speech Recognition

“…Zeghidour et al [24] proposed an implementation with with Gabor filters and Ravanelli et al [25] with rectangular filters (SincNet). Other work on trainable filterbanks includes that of Seki et al [26], who proposed an architecture based on a filter layer combined with a DNN where the filter features were directly computed with a log-compression after the filter layer. In that study the gain, central frequency, bandwidth and filter shape were free to train, whilst in SincNet only two parameters are free to train, defining the filters in the first layer.…”