Switching Linear Dynamical Systems for Noise Robust Speech Recognition

Abstract

Abstract. Real world applications such as hands-free dialling in cars may have to deal with potentially very noisy environments. Existing state-of-the-art solutions to this problem use featurebased HMMs, with a preprocessing stage to clean the noisy signal. However, the effect that raw signal noise has on the induced HMM features is poorly understood, and limits the performance of the HMM system. An alternative to feature-based HMMs is to model the raw signal, which has the potential advantage that including an explicit noise model is straightforward. Here we jointly model the dynamics of both the raw speech signal and the noise, using a Switching Linear Dynamical System (SLDS). The new model was tested on isolated digit utterances corrupted by Gaussian noise. Contrary to the Autoregressive HMM and its derivatives, which provides a model of uncorrupted raw speech, the SLDS is comparatively noise robust and also significantly outperforms a state-of-the-art feature-based HMM. The computational complexity of the SLDS scales exponentially with the length of the time series. To counter this we use Expectation Correction which provides a stable and accurate linear-time approximation for this important class of models, aiding their further application in acoustic modelling.

A unified model for context-based behavioural modelling and classification

“…Various problems have been successfully solved using the SLDS. These include speech recognition (Mesot and Barber, 2007), econometric (Kim, 1994) and human figure tracking problems . Several other human behaviour detection applications are discussed in Turaga et al (2008).…”

Maritime piracy situation modelling with dynamic Bayesian networks

“…These include the switching Kalman filter and the switching state space model [19]. The SLDS has been successfully applied to various problems that include human motion modelling in computer vision [21], econometrics [22] and speech recognition [23]. No attempts to use SLDS as a generative model for data synthesis have been found in literature.…”

A Bayesian Approach to Switching Linear Gaussian State-Space Models for Unsupervised Time-Series Segmentation

“…1 A similar approximation approach to a constrained non-fully Bayesian switching LGSSM has independently been introduced in [9] in the context of supervised speech processing.…”

“…Segmentation is then be performed based on an application of Bayes' rule to infer which set of parameters was most likely to have generated the observations at any particular time. This approach has successfully been used in several application domains such as finance, speech processing, modeling of human motion and medicine [1,10,11,9]. In our fully unsupervised scenario, the underlying number of different parameter sets is not known in advance and needs to be estimated.…”