Modified Kanerva model. Results for real time word recognition

This paper describes results obtained using the modified Kanerva model to perform word recognition in continuous speech after being trained on the multi-speaker Alvey 'Hotel' speech corpus. Theoretical discoveries have recently enabled us to increase the speed of execution of part of the model by two orders of magnitude over that previously reported by Prager & Fallside. The memory required for the operation of the model has been similarly reduced. The recognition accuracy reaches 95% without syntactic constraints when tested on different data from seven trained speakers. Real time simulation of a model with 9,734 active units is now possible in both training and recognition modes using the Alvey PARSIFAL transputer array. The modified Kanerva model is a static network consisting of a fixed nonlinear mapping (location matching) followed by a single layer of conventional adaptive links. A section of preprocessed speech is transformed by the non-linear mapping to a high dimensional representation. From this intermediate representation a simple linear mapping is able to perform complex pattern discrimination to form the output, indicating the nature of the speech features present in the input window.

Water quality of two semi-closed areas in the United Arab Emirates Coast along the Arabian Gulf:a comparative study

A comparative study was carried out between the two biggest creeks along the Arabian Gulf coast of the United Arab Emirates to evaluate impacts of sewage and industrial effluents on their hydrochemical characteristics. Surface and bottom water samples were collected from Abu Dhabi and Dubai creeks during the period from October 1994 to September 1995. The hydrochemical parameters studied were: temperature (21.10-34.00°C), salinity (37.37-47.09%), transparency (0.50-10.0 m), pH (7.97-8.83), dissolved oxygen (1.78-13.93 mg/l) and nutrients ammonia (ND- 13.12,ug-at N/1), nitrite (ND-6.66 ,ug-at N/1), nitrate (ND- 41.18 ,ug-at N/1), phosphate (ND- 13.06 ,ug-at P/1), silicate (0.68-32.50 ,ug-at Si/1), total phosphorus (0.26- 21.48 ,ug-at P/1), and total silicon (0.95- 40.32 ,ug-at Si/1). The present study indicates clearly that seawater of Abu-Dhabi Creek was warmer (28.l2°C) than Dubai (27.56°C) resulting in a higher rate of evaporation. Owing to more evaporation, salinity levels showed higher levels at Abu Dhabi (43.33%) compared to Dubai (39.03%) seawater. The study also revealed higher secchi disc readings at Abu Dhabi Creek (4.68 m) as compared to Dubai Creek (2.60 m) suggesting more transparency at Abu Dhabi Creek. Whereas, seawater of Dubai exhibited higher levels of pH (1.03 times), and dissolved oxygen (1.05 times) than Abu Dhabi seawater due to an increase in productivity. Meantime, seawater of Dubai showed higher tendency to accumulate ammonia (8.22 times), nitrite (10.93 times), nitrate (5.85 times), phosphate (10.64 times), silicate (1.60 times), total phosphorus (3.19 times), and total silicon (1.54 times) compared to Abu Dhabi seawater due to the enrichment of seawater at Dubai with domestic sewage waters which has distinctly elevated the levels of the nutrient salts particularly in inner-most parts of the creek leading to eutrophication signs. The changes occurred in the receiving creek water of Dubai as a result of waste-water disposal that have also reflected on the atomic ratios of nit: Effect of pollution rogen: phosphorus: silicon.

Nutrient salts in the United Arab Emirates waters (the Arabian Gulf and the Gulf of Oman)

This paper deals with the levels and distributions of nutrient salts in the United Arab Emirates waters. Water samples were collected bimonthly during 1994-1995 from the marine environment of the United Arab Emirates, which extends more than 800km along the Arabian Gulf and the Gulf of Oman. Concentrations of ammonium, nitrite, nitrate, phosphate, silicate, as well as total concentrations of total dissolved nitrogen, phosphorus, and silicon in the area were scattered in the ranges: (ND-6.32; mean: 0.84 µg-at N/l), ND-3.02; mean: 0.42 µg-at N/l), (ND-10.88; mean: 1.18 µg-at N/1), (ND-4.22; mean: 0.62 µg-at P/l), (1.14-28.80; mean: 6.52 µg-at Si/l), (1.52-39.58; mean: 12.28 µg-at N/l), (0.40-4.98; mean: 1.07 µg-at P/l), and (2.77-44.74; mean: 13.02 Si/l) respectively. Of inorganic nitrogen species, ammonium was the highest in the Arabian Gulf waters and nitrate was the highest at the Gulf of Oman. The dissolved inorganic nitrogen total species, phosphate and silicate amounted to 16.4, 47.6, 56.5% respectively, of the concentrations of nitrogen, phosphorus and silicon in the Arabian Gulf and 22.6, 64.4, 44.9% respectively, in the Gulf of Oman, indicating that more than 80% of nitrogen was present in organic forms. Distributions of nutrient in the two regions were higher in the summer season and lower in the winter season due to the oxidation of organic materials. Regional distributions revealed higher values for nitrite (1.3 times), nitrate (2.8 times), phosphate (2.2 times), total dissolved nitrogen (1.3 times), total dissolved phosphorus (1.6 times), and total dissolved silicon (1.3 times) in the Gulf of Oman compared to the Arabian Gulf, indicating more oligotrophic conditions at the Arabian Gulf Whereas no distinct patterns of distribution were observed in the Arabian Gulf waters, an increase in the seaward direction was measured at the Gulf of Oman. Vertical distributions indicated a general increase with depth in the two regions. The mean ratios for total concentrations of phosphorus, nitrogen, and silicon in the Arabian Gulf (1: 11.6: 12.6) and the Gulf of Oman (1: 10.1: 11.8) were lower than the Redfield ratio.

Word boundary modelling and full covariance gaussians for Arabic Speech-to-Text systems

This paper describes recent improvements to the Cambridge Arabic Large Vocabulary Continuous Speech Recognition (LVCSR) Speech-to-Text (STT) system. It is shown that wordboundary context markers provide a powerful method to enhance graphemic systems by implicit phonetic information, improving the modelling capability of graphemic systems. In addition, a robust technique for full covariance Gaussian modelling in the Minimum Phone Error (MPE) training framework is introduced. This reduces the full covariance training to a diagonal covariance training problem, thereby solving related robustness problems. The full system results show that the combined use of these and other techniques within a multi-branch combination framework reduces the Word Error Rate (WER) of the complete system by up to 5.9% relative. Copyright © 2011 ISCA.

IPLR: An online resource for Greek word-level and sublexical information

We present a new online psycholinguistic resource for Greek based on analyses of written corpora combined with text processing technologies developed at the Institute for Language & Speech Processing (ILSP), Greece. The "ILSP PsychoLinguistic Resource" (IPLR) is a freely accessible service via a dedicated web page, at http://speech.ilsp.gr/iplr. IPLR provides analyses of user-submitted letter strings (words and nonwords) as well as frequency tables for important units and conditions such as syllables, bigrams, and neighbors, calculated over two word lists based on printed text corpora and their phonetic transcription. Online tools allow retrieval of words matching user-specified orthographic or phonetic patterns. All results and processing code (in the Python programming language) are freely available for noncommercial educational or research use. © 2010 Springer Science+Business Media B.V.

Discriminative spoken language understanding using word confusion networks

Current commercial dialogue systems typically use hand-crafted grammars for Spoken Language Understanding (SLU) operating on the top one or two hypotheses output by the speech recogniser. These systems are expensive to develop and they suffer from significant degradation in performance when faced with recognition errors. This paper presents a robust method for SLU based on features extracted from the full posterior distribution of recognition hypotheses encoded in the form of word confusion networks. Following [1], the system uses SVM classifiers operating on n-gram features, trained on unaligned input/output pairs. Performance is evaluated on both an off-line corpus and on-line in a live user trial. It is shown that a statistical discriminative approach to SLU operating on the full posterior ASR output distribution can substantially improve performance both in terms of accuracy and overall dialogue reward. Furthermore, additional gains can be obtained by incorporating features from the previous system output. © 2012 IEEE.

Using sub-word-level information for confidence estimation with conditional random field models

The task of word-level confidence estimation (CE) for automatic speech recognition (ASR) systems stands to benefit from the combination of suitably defined input features from multiple information sources. However, the information sources of interest may not necessarily operate at the same level of granularity as the underlying ASR system. The research described here builds on previous work on confidence estimation for ASR systems using features extracted from word-level recognition lattices, by incorporating information at the sub-word level. Furthermore, the use of Conditional Random Fields (CRFs) with hidden states is investigated as a technique to combine information for word-level CE. Performance improvements are shown using the sub-word-level information in linear-chain CRFs with appropriately engineered feature functions, as well as when applying the hidden-state CRF model at the word level.