Noel Massey

The uses of social television

We present the results of two studies on social television concepts. In one study, a social TV prototype was tested in the field, allowing groups of users watching television at home to talk to each other over an audio link. Specific patterns of use are described, showing that users did perceive the system as valuable. In another study, focus groups were presented with several social TV concepts, and their responses were collected. These participants saw only moderate to marginal value in the concept. We discuss the discrepancy with reference to the limitations of each method. Based on our analysis, we conclude that our social television experience provides user value under certain favorable conditions. Participants deal with potential conflicts between conversation and television audio without the need for additional technical support, and there is no indication that a video link would improve the experience. However, designing for the social dynamics at the beginnings, ends, and outside of conversations remains an open challenge.

Driver inattention detection through intelligent analysis of readily available sensors

Driver inattention is estimated to be a significant factor for over 25% of all crashes. A system that could accurately detect driver inattention could aid in reducing this number. In contrast to using specialized sensors or video cameras to monitor the driver we detect driver inattention by using only readily available sensors. A classifier was trained using collision avoidance systems (CAS) sensors, which was able to accurately identify 80% of driver inattention and could be added to a vehicle without incurring the cost of additional sensors.

Perceptions of value: the uses of social television

We present the results of two studies on social television concepts. In one study, a Social TV prototype was tested in the field, allowing groups of users watching television at home to talk to each other over an audio link. Specific patterns of use are described, showing that users did perceive the system to be valuable. In another study, focus groups were presented with several Social TV concepts, and their responses were collected. These participants saw only moderate to marginal value in the concept. We discuss the discrepancy with reference to the limitations of each method.

A high quality text-to-speech system composed of multiple neural networks

While neural networks have been employed to handle several different text-to-speech tasks, ours is the first system to use neural networks throughout, for both linguistic and acoustic processing. We divide the text-to-speech task into three subtasks, a linguistic module mapping from text to a linguistic representation, an acoustic module mapping from the linguistic representation to speech, and a video module mapping from the linguistic representation to animated images. The linguistic module employs a letter-to-sound neural network and postlexical neural network. The acoustic module employs a duration neural network and a phonetic neural network. The visual neural network is employed in parallel to the acoustic module to drive a talking head. The use of neural networks that can be retrained on the characteristics of different voices and languages affords our system a degree of adaptability and naturalness heretofore unavailable.

Investigating the Use of Voice and Text Chat in a Social Television System

We tested a social television system with both voice and text communication in the field for one week, in four households. Participants expressed a preference for the freeform communication features over an earlier system with restricted communication options. Use of the communication features was nevertheless sporadic. Text was used more often than voice communication. However, both were usually used in combination. The in-home context had marked effects on the usage patterns, and we observed issues of privacy and identity verification.

Detecting driver inattention in the absence of driver monitoring sensors

A classifier was trained to detect driver inattention using output from typical sensors available on modern vehicles equipped with a Collision Avoidance System (CAS). A driving simulator was used to collect driver and vehicle data from ten subjects during normal driving periods and during periods where drivers looked away from their forward view as they turned their head to look at a detailed image in their blindspot. This data was used to train a 2-state (attentive, inattentive) and 3-state (attentive, inattentive left, inattentive right) classifier. Monitoring of driver attention is typically performed by in-vehicle cameras or specialized sensors but we were able to use simulated CAS sensor output to detect over 80% of driver inattention time segments in the absence of driver monitoring sensors.

An architecture for an intelligent driver assistance system

We are building an adaptive driver support system using an agent architecture and machine learning techniques. The goal of the system is to help the drivers have a safer, more enjoyable and more productive driving experiences, by managing their attention and workload. In this paper, we describe the overall architecture of the driver support system and how we apply machine learning techniques to have the system adapt to the driving behavior of each individual driver. The architecture has been partially implemented in a prototype system built upon a high-fidelity driving simulator, allowing us to run experimental tests on the interaction between the system and human users. Once the system demonstrates the desired capabilities, it will be tested in a real car in an actual driving environment.

Transition-based speech synthesis using neural networks

Prior attempts to use neural networks to synthesize speech from a phonetic representation have used the neural network to generate a frame of input to a vocoder. As this requires the neural network to compute one output for each frame of speech from the vocoder, this can be computationally expensive. An alternative implementation is to model the speech as a series of gestures, and let the neural network generate parameters describing the transitions of the vocoder parameters during these gestures. Experiments have shown that acceptable speech quality is produced when each gesture is half of a phonetic segment and the transition model is a set of cubic polynomials describing the variation of each vocoder parameter during the gesture. This results in a significant reduction in computational cost.

Toward modeling and classification of naturalistic driving

Intelligent systems in automobiles need to be aware of the driving and driver context Available sensor data streams have to be modeled and monitored in order to do so. We describe a machine learning approach to accomplish this through collecting a large database of naturalistic driving data in a driving simulator. Preliminary experiments with a smaller dataset indicate successful modeling of naturalistic driving with hierarchical sequential models such as hidden Markov models.

Transients at stop‐consonant releases

The acoustic properties and perceptual significance of transients at the release of stop consonants and affricates was studied. The theory of transient production is based on a model in which pressure is built up in a closed tube and the radiated sound is calculated following an abrupt release of the pressure. Acoustic analysis of the amplitude and frequency content of transients in utterances from ten speakers was used as a guide in setting the parameters of the model. A speech synthesizer was modified to allow the generation of transients with frequency content similar to that observed in natural speech. Utterances consisting of a stop consonant or affricate followed by the vowel /a/, with various amplitudes of an initial transient, were synthesized, and listeners were asked to judge the naturalness of the stimuli. Listeners tended to prefer the presence of a transient with an amplitude approximating the measured amplitude in natural speech, although there was some variability in the responses. [Work su...