Showing papers by "Arnon D. Cohen published in 1995"

Unsupervised text independent speaker classification

Abstract: Speaker recognition and verification has been used in a variety of commercial, forensic and military applications. The classical problem is that of supervised recognition, in which there is sufficient a priori information on the speakers to be identified. In such cases, the recognition system has speaker models, estimated during training sessions. This paper deals with the problem of unsupervised speaker classification, where no a priori speaker information is available. The algorithm accepts multi-speaker dialogue speech data, estimates the number of speakers and assigns each speech segment to its speaker. Preliminary results are described.

CA 15-3, mucin-like carcinoma-associated antigen and tissue polypeptide-specific antigen: correlation to disease state and prognosis in breast cancer patients.

Abstract: Serum testing for the tumor markers CA 15-3 and mucin-like carcinoma-associated antigen (MCA) was performed in 144 patients with breast cancer. Of those patients, 127 were also tested for a third tumor marker, tissue polypeptide-specific antigen (TPS). At the time of testing, 73 patients (51%) were without evidence of disease and 71 patients (49%) had active disease. Mean follow-up time was 16 months. Positivity rates were: 29.8% for CA 15-3, 47.2% for MCA, and 58.3% for TPS. Mean value for patients with no evidence of disease as compared to patients with active disease was 21 U/ml and 117 U/ml for CA 15-3 (P = 0.001), 10.2 kU/l and 22.6 kU/l for MCA (P = 0.001), and 148 U/l and 310 U/l for TPS (P = 0.02). Two year actuarial survival, according to values below and above an established cutoff for CA 15-3 were 80% and 31% (P < 0.001), 75% and 45% for MCA (P < 0.01), and 82% and 50% for TPS (P < 0.01). We conclude that CA 15-3, MCA and TPS values reflect disease state and prognosis in breast cancer patients.

HMM phoneme recognition with supervised training and Viterbi algorithm

Abstract: An HMM continuous Hebrew phoneme recognition system, that requires no manual segmentation for its training was developed. A relatively small Hebrew data base was acquired for training and recognition of phonemes in continuous speech. One of the main problems in phoneme recognition, that of manual segmentation of the training data base, was overcome by a special training algorithm. The Viterbi algorithm was used in the recognition stage, and the evaluation of the results was done with the Levenshtein distance measure. Initial recognition results of Hebrew phonemes for speaker independent, text dependent cases were 69.4% correct phoneme recognition.

Monitoring of separate lung ventilation in anesthesia and intensive care

Abstract: Analysis of lung sounds for monitoring and diagnosis of pulmonary function is well known. The application of these techniques to ventilation monitoring during anesthesia or in intensive care, may prove a convenient, accurate and very low latency monitoring device. A prototype of such a system is described. The system is able to monitor the current state of ventilation and automatically classify among three types: Tracheal Intubation (TRI), Left Lung Intubation (LLI) and Right Lung Intubation (RLI), Preliminary experiments, performed on dogs, show mean classification accuracy of 90%.