André Gustavo Adami

The SuperSID project: exploiting high-level information for high-accuracy speaker recognition

The area of automatic speaker recognition has been dominated by systems using only short-term, low-level acoustic information, such as cepstral features. While these systems have indeed produced very low error rates, they ignore other levels of information beyond low-level acoustics that convey speaker information. Recently published work has shown examples that such high-level information can be used successfully in automatic speaker recognition systems and has the potential to improve accuracy and add robustness. For the 2002 JHU CLSP summer workshop, the SuperSID project (http://www.clsp.jhu.edu/ws2002/groups/supersid/) was undertaken to exploit these high-level information sources and dramatically increase speaker recognition accuracy on a defined NIST evaluation corpus and task. The paper provides an overview of the structure, data, task, tools, and accomplishments of this project. Wide ranging approaches using pronunciation models, prosodic dynamics, pitch and duration features, phone streams, and conversational interactions were explored and developed. We show how these novel features and classifiers indeed provide complementary information and can be fused together to drive down the equal error rate on the 2001 NIST extended data task to 0.2% - a 71% relative reduction in error over the previous state of the art.

Modeling prosodic dynamics for speaker recognition

Most current state-of-the-art automatic speaker recognition systems extract speaker-dependent features by looking at short-term spectral information. This approach ignores long-term information that can convey supra-segmental information, such as prosodics and speaking style. We propose two approaches that use the fundamental frequency and energy trajectories to capture long-term information. The first approach uses bigram models to model the dynamics of the fundamental frequency and energy trajectories for each speaker. The second approach uses the fundamental frequency trajectories of a predefined set of words as the speaker templates and then, using dynamic time warping, computes the distance between the templates and the words from the test message. The results presented in this work are on Switchboard I using the NIST Extended Data evaluation design. We show that these approaches can achieve an equal error rate of 3.7%, which is a 77% relative improvement over a system based on short-term pitch and energy features alone.

Medication Adherence in Healthy Elders: Small Cognitive Changes Make a Big Difference

Objective: This was a cross-sectional study of the ability of independently living healthy elders to follow a medication regimen. Participants were divided into a group with High Cognitive Function (HCF) or Low Cognitive Function (LCF) based on their scores on the ADAS-Cog. Method: Thirty-eight participants aged 65 or older and living independently in the community followed a twice-daily vitamin C regimen for 5 weeks. Adherence was measured using an electronic 7-day pillbox. Results: The LCF group had significantly poorer total adherence than the HCF group (LCF: 63.9 ± 11.2%, HCF: 86.8 ± 4.3%, t 36 = 2.57, p = .007), and there was a 4.1 relative risk of non-adherence in the LCF group as compared to the HCF group. Discussion: This study has important implications for the conduct of clinical drug trials, as it provides strong evidence that even very mild cognitive impairment in healthy elderly has a detrimental impact on medication adherence.

Unobtrusive assessment of activity patterns associated with mild cognitive impairment

Timely detection of early cognitive impairment is difficult. Measures taken in the clinic reflect a single snapshot of performance that might be confounded by the increased variability typical in aging and disease. We evaluated the use of continuous, long‐term, and unobtrusive in‐home monitoring to assess neurologic function in healthy and cognitively impaired elders.

An Electronic Pillbox for Continuous Monitoring of Medication Adherence

We have developed an instrumented pillbox, called a MedTracker, which allows monitoring of medication adherence on a continuous basis. This device improves on existing systems by providing mobility, frequent and automatic data collection, more detailed information about non-adherence and medication errors, and the familiar interface of a 7-day drug store pillbox. We report on the design of the MedTracker, and on the results of a field trial in 39 homes to evaluate the device

Salient EEG Channel Selection in Brain Computer Interfaces by Mutual Information Maximization

Modern brain computer interface (BCI) applications use information obtained from the users electroencephalogram (EEG) to estimate the mental states. Selecting an optimal subset of the EEG channels instead of using all of them is especially important for ambulatory EEG where the user is mobile due to reduced data communication and computational load requirements. In addition, elimination of irrelevant sensors improves the robustness of the classification system by reducing dimensionality. In this paper, we propose a filter approach for EEG channel selection using mutual information (MI) maximization. This method ranks the EEG channels, such that the MI between the selected sensors and class labels is maximized. This selection criterion is known to reduce classification error. We employ a computationally efficient approach for MI estimation and EEG channel ranking. This approach is illustrated on EEG data recorded from three subjects performing two mental tasks. Experiment results show that the proposed approach works well and the position of the selected channels using the proposed method is consistent with the expected cortical areas for the mental tasks

Distributed Healthcare: Simultaneous Assessment of Multiple Individuals

Employing pervasive computing technologies can help enable continuous patient monitoring and assessment in various settings outside of hospitals, lowering healthcare costs and allowing earlier detection of problems

The 2004 MIT Lincoln Laboratory speaker recognition system

The MIT Lincoln Laboratory submission for the 2004 NIST speaker recognition evaluation (SRE) was built upon seven core systems using speaker information from short-term acoustics, pitch and duration prosodic behavior, and phoneme and word usage. These different levels of information were modeled and classified using Gaussian mixture models, support vector machines and N-gram language models and were combined using a single layer perceptron fuser. The 2004 SRE used a new multi-lingual, multi-channel speech corpus that provided a challenging speaker detection task for the above systems. We describe the core systems used and provide an overview of their performance on the 2004 SRE detection tasks.

Channel selection and feature projection for cognitive load estimation using ambulatory EEG

We present an ambulatory cognitive state classification system to assess the subjects mental load based on EEG measurements. The ambulatory cognitive state estimator is utilized in the context of a real-time augmented cognition (AugCog) system that aims to enhance the cognitive performance of a human user through computer-mediated assistance based on assessments of cognitive states using physiological signals including, but not limited to, EEG. This paper focuses particularly on the offline channel selection and feature projection phases of the design and aims to present mutual-information-based techniques that use a simple sample estimator for this quantity. Analyses conducted on data collected from 3 subjects performing 2 tasks (n-back/Larson) at 2 difficulty levels (low/high) demonstrate that the proposed mutual-information-based dimensionality reduction scheme can achieve up to 94% cognitive load estimation accuracy.

Unobtrusive assessment of mobility.

This paper describes a model-based approach to the unobtrusive monitoring of elders in their home environment to assess their health, daily activities, and cognitive function. We present a semi-Markov model representation with automated learning to characterize individual elders mobility in the home environment. The assessed mobility can be used to characterize the elders speed of walking and can serve as one of the predictors of future cognitive functionality and the ability of elders to live independently in their home environment