Robust and quality-assured audio-visual communication

点击翻译按钮获取中文摘要

Auditory perceptual modeling of distortions plays a vital role in enabling today's widespread utilization of compressed speech and audio. Every voice conversation over a cellular telephone or download of MP3 music entails the use of perceptual-model based measurement of distortions.  When a speech or audio signal is compressed into a digital format, a distortion model is used to determine where in the signal to "hide" distortions to make them least perceptible. We propose to research using a new generation of perceptual models to build more powerful tools for speech communication and audio distribution over emerging and future network configurations.   One goal is to build machine intelligence to enable computers to rate the quality of a speech signal in the same manner as a panel of human listeners.  This would enable communication network elements and electronic devices to better optimize resource usage in accordance with human-perceived speech quality. Another goal is to devise new digital formatting algorithms that would enable efficient and reliable distribution of audio to a very large number of clients. The format will be very flexible so that the user-perceived audio quality at each client is the best possible given the particular quality of the client's network connection.  One anticipated benefit is that high-fidelity audio net-casting can reach a much greater number of users.