Evaluating and improving morpho-syntactic classification over multiple corpora using pre-trained, “off-the-shelf”, parts-of-speech tagging tools

his paper evaluates six commonly available parts-of-speech tagging tools over corpora other than those upon which they were originally trained. In particular this investigation measures the performance of the selected tools over varying styles and genres of text without retraining, under the assumption that domain specific training data is not always available. An investigation is performed to determine whether improved results can be achieved by combining the set of tagging tools into ensembles that use voting schemes to determine the best tag for each word. It is found that while accuracy drops due to non-domain specific training, and tag-mapping between corpora, accuracy remains very high, with the support vector machine-based tagger, and the decision tree-based tagger performing best over different corpora. It is also found that an ensemble containing a support vector machine-based tagger, a probabilistic tagger, a decision-tree based tagger and a rule-based tagger produces the largest increase in accuracy and the largest reduction in error across different corpora, using the Precision-Recall voting scheme.

Distributed audio network for speech enhancement in challenging noise backgrounds

This paper presents a new approach to enhance speech based on a distributed microphone network. Each microphone is used to simultaneously classify the input into either one of the noise types or as speech. For enhancing the speech signal a modified spectral subtraction approach is used that utilise the sound information of the entire network to update the noise model even during speech. This improves the reduction of the ambient noise, especially for non-stationary noise types such as street or beach noise. Experiments demonstrate the effectiveness of the proposed system.

Adaptive speech enhancement with varying noise backgrounds

We present a new approach for speech enhancement in the presence of non-stationary and rapidly changing background noise. A distributed microphone system is used to capture the acoustic characteristics of the environment. The input of each microphone is then classified either as speech or one of the predetermined noise types. Further enhancement of speech in respective microphones is carried out using a modified spectral subtraction algorithm that incorporates multiple noise models to quickly adapt to rapid background noise changes. Tests on real world speech captured under diverse conditions demonstrate the effectiveness of this method.