Microphone-Independent Robust Signal Processing Using Probabilistic Optimum Filtering

A new mapping algorithm for speech recognition relates the features of simultaneous recordings of clean and noisy speech. The model is a piecewise nonfinear transformation appfied to the noisy speech feature. The transformation is a set of multidimensional linear least-squares filters whose outputs are combined using a conditional Gaussian model. The algorithm was tested using SRI's DECIPHER TM speech recognition system [1-5]. Experimental results show how the mapping is used to reduce recognition errors when the training and testing acoustic environments do not match. 1. I N T R O D U C T I O N In many practical situations an automatic speech recognizer has to operate in several different but well-defined acoustic environments. For example, the same recognition task may be implemented using different microphones or transmission channels. In this situation it may not be practical to recollect a speech corpus to train the acoustic models of the recognizer. To alleviate this problem, we propose an algorithm that maps speech features between two acoustic spaces. The models of the mapping algorithm are trained using a small database recorded simultaneously in both environments. In the case of steady-state additive homogenous noise, we can derive a MMSE estimate of the clean speech filterbank-log energy features using a model for how the features change in the presence of this noise [6-7]. In these algorithms, the estimated speech spectrum is a function of the global spectral signal-to-noise ratio (SNR), the instantaneous spectral SNR, and the overall spectral shape of the speech signal. However, after studying simultaneous recordings made with two microphones, we befieve that the relationship between the two simultaneous features is nonlinear. We therefore propose to use a piecewise-nonlinear model to relate the two feature spaces. 1.1. Related Work on Feature Mapping Several algorithms in the literature have focused on experimentally training a mapping between the noisy features and the clean features [8-13]. The proposed algorithm differs from previous algorithms in several ways: • The MMSE estimate of the clean speech features in noise is trained experimentally rather than with a model as in [6, 7]. • Several frames are joined together similar to [13]. • The conditional PDF is based on a generic noisy feature not necessarily related to the feature that we are trying to estimate. For example, we could condition the estimate of the cepstral energy on the instantaneous spectral SNR vector. • Multidimensional least-squares filters are used for the mapping transformation. This exploits the correlation of the features over time and among components of the spectral ,features at the same time. • Linear transformations are combined together without hard decisions. • All delta parameters are computed after mapping the cepstrum and cepstral energy. • The mapping parameters are trained using stereo recordings with two different microphones. Once trained, the mapping parameters are fixed. • The algorithm can either map noisy speech features to clean features during training, or clean features to noisy features during recognition. 1.2. Related W o r k on Adapta t ion The algorithm used to map the incoming features into a more robust representation has some similarities to work on model adaptation. Some of the high-level differences between hidden Markov model (HMM) adaptation and the mapping algorithms proposed in this paper are: • The mapping algorithm works by primarily correcting shifts in the mean of the feature set that are correlated with observable information. Adapting HMM model parameters has certain degrees of freedom that the mapping algorithm does not havefor example the ability to change state variances, and mixture weights. • Two HMM states that have identical probability distributions and are not tied can have different distributions after adaptation. These distributions cannot be differentiated by mapping features. • The mapping algorithms described in this paper are able to incoiporate many pieces of information that have been tra-