Robust continuous speech recognition using parallel model combination

Abstract

This paper addresses the problem of automatic speech recognition in the presence of interfering noise. It focuses on the Parallel Model Combination (PMC) scheme, which has been shown to be a powerful technique for achieving noise robustness. Most experiments reported on PMC to date have been on small, 10-50 word vocabulary systems. Experiments on the Resource Management (RM) database, a 1000 word continuous speech recognition task, reveal compensation requirements not highlighted by the smaller vocabulary tasks. In particular, that it is necessary to compensate the dynamic parameters as well as the static parameters to achieve good recognition performance. The database used for these experiments was the RM speaker independent task with either Lynx Helicopter noise or Operation Room noise from the NOISEX-92 database added. The experiments reported here used the HTK RM recogniser developed at CUED modi ed to include PMC based compensation for the static, delta and delta-delta parameters. After training on clean speech data,the performance of the recogniser was found to be severely degraded when noise was added to the speech signal at between 10dB and 18dB. However, using PMC the performance was restored to a level comparable with that obtained when training directly in the noise corrupted environment.

Investigation and Development of Methods for Improving Robustness of Automatic Speech Recognition Algorithms in Complex Acoustic Environments

“…The same linearized distortion model is used to modify features distribution parameterized as a GMMs (Gaussian Mixture Model). A number of other successful approaches to adapt GMM-based acoustic models were also developed such as MLLR, CMLLR (fMLLR) [16], MAP [17], PMC [18], and some their combinations.…”

Parallel model combination and word recognition in soccer audio

“…Of these methods only the resampling has been applied to delta and acceleration features. The resampling PMC [6] was used for the model combination. To recalculate the parameters, three separate equations are used.…”

“…However, linear features are rare in audio recognition systems, with nonlinear features such as MFCCs and PLPs used, which compress the audio spectrum for more robust recognition. The compression causes the exact model combination through numerical integration to be computationally intensive, for this reason several approximations have been tried and PMC schemes [6] have been shown to be successful for the improvement of speech recognition in noisy environments.…”

“…To tackle such a problem, three well known methods for dealing with noisy speech when using an HMM structure can be used and depending on the level of information that is known about the noise stream a different model may be appropriate. The methods are multiconditional training, parallel model combination (PMC) [6] and the factorial HMM (FHMM) [7]. Multi-conditional training is simply training the speech models with noisy conditions.…”

The synergy between bounded-distance HMM and spectral subtraction for robust speech recognition

“…PMC (Gales and Young (1996)) and MLLR (Gales and Woodland (1996)) are good examples of methods within this category.…”