The ability to predict the Remaining Useful Life (RUL) of Rotating Machines is a highly desirable function of Automated Condition Monitoring (ACM) systems. Typically, vibration signals are acquired through contact with the machine and used for monitoring. In this paper, a novel implementation of the ubiquitous feature extraction approach Envelope Analysis (EA) is applied to acoustic noise signals (< 25 kHz) to predict the RUL of a rotating machine. A well known drawback of the EA approach is that the frequency band of interest must be known or pre-estimated. Therefore, this approach is compared to a Data-Driven approach to feature extraction which utilizes an Information Theoretic approach to feature selection that does not require any a-priori information regarding the frequency band of interest. It is shown that the Data- Driven approach, with an accuracy of 97.7%, significantly outperforms the EA approach, with an accuracy of 93.7%. This study also shows that the improved performance of the Data-Driven approach is due to new information being uncovered in spectral locations across the entire spectrum from 0 to 25 kHz, and not just within one frequency band typically used by the EA approach.

We present a novel technique for radio transmitter identification based on frequency domain characteristics. Our technique detects the unique features imbued in a signal as it passes through a transmit chain. We are the first to propose the use of discriminatory classifiers based on steady state spectral features. In laboratory experiments, we achieve 97% accuracy at 30 dB SNR and 66% accuracy at OdB SNR based on eight identical universal software radio peripherals (USRP) transmitters. Our technique can be implemented using today's low cost high-volume receivers and requires no manual performance tuning.

In phoneme recognition experiments, it was found that approximately 75% of misclassified frames were assigned labels within the same broad phonetic group (BPG). While the phoneme can be described as the smallest distinguishable unit of speech, phonemes within BPGs contain very similar characteristics and can be easily confused. However, different BPGs, such as vowels and stops, possess very different spectral and temporal characteristics. In order to accommodate the full range of phonemes, acoustic models of speech recognition systems calculate input features from all frequencies over a large temporal context window. A new phoneme classifier is proposed consisting of a modular arrangement of experts, with one expert assigned to each BPG and focused on discriminating between phonemes within that BPG. Due to the different temporal and spectral structure of each BPG, novel feature sets are extracted using mutual information, to select a relevant time-frequency (TF) feature set for each expert. To construct a phone recognition system, the output of each expert is combined with a baseline classifier under the guidance of a separate BPG detector. Considering phoneme recognition experiments using the TIMIT continuous speech corpus, the proposed architecture afforded significant error rate reductions up to 5% relative

Image transforms, such as the discrete cosine, are widely used to extract visual features from the speaker's mouth region to be used in automatic speechreading and audio-visual speech recognition. Typically, the spatial frequency components with the highest energy in the transform space are retained for recognition. This paper proposes an alternative technique for selecting such features, by utilizing the mutual information criterion instead. Mutual information between each individual spatial frequency component and the speech classes of interest is employed as a measure of its appropriateness for speech classification. The highest mutual information components are then selected as visual speech features. Extensions to this scheme by using joint mutual information between candidate feature pairs and classes are also considered. The algorithm is tested on visual-only speech recognition of connected-digit strings, using an appropriate audio-visual database. For low-dimensional visual feature vectors, the proposed method significantly outperforms features selected by means of energy, reducing word error rate by as much as 20% relative. These gains however diminish as higher feature dimensionalities are allowed.

Information concerning the identity of subword units such as phones cannot easily be pinpointed because it is broadly distributed in time and frequency.Continuing earlier work, we use Mutual Information as measure of the usefulness of individual time-frequency cells for various speech classification tasks, using the hand-annotations of the TIMIT database as our ground truth.Since different broad phonetic classes such as vowels and stops have such different temporal characteristics, we examine mutual information separately for each class, revealing structure that was not uncovered in earlier work; further structure is revealed by aligning the time-frequency displays of each phone at the center of their hand-marked segments, rather than averaging across all possible alignments within each segment.Based on these results, we evaluate a range of vowel classifiers over the TIMIT test set and show that selecting input features according to the mutual information criteria can provides a significant increase in classification accuracy.

Speechreading is the ability to understand a speaker’s thoughts by watching the movements of the face and body and by using the information provided by the situation and the language. People with normal hearing and the hearing impaired use speechreading to augment communication especially in noisy environments. Just as people learn this skill, machines can be trained to understand a speakers meaning. Audio-Visual Automatic Speech Recognition (AV ASR) systems use audio and visual information to recognize what has been ‘said’. The speech sounds and movements provided need not be standard speech sounds or movements. The system will provide recognition given audio information only, visual information only or both.

Audio-visual automatic speech recognition systems use visual information to enhance ASR systems in clean and noisy environments. This paper investigates a number of different visual feature extraction methods. It was observed that when performing visual speech recognition the visual feature vector requires a base level of detail for improved recognition. Geometric feature extraction provides lower recognition than pixel based methods due to the loss of characteristic speech information such as protrusion etc. Downsampling of images reduces visual recognition scores due to the loss of detail in the images. Also, the role of dynamic features was investigated for improved recognition. It was observed that static features alone outperform a combination of both static and dynamic features when restricting the dimension of the feature vector e.g. 50. This illustrates that the need for a certain level of detail in visual speech recognition is a higher priority than dynamic information. Once this base level of detail is attained the dynamic features should then be able to improve the recognition rate.