Voice extraction by on-line signal separation and recovery

The paper presents a formulation and an implementation of a system for voic e output extraction (VOX) in real-time and near-real-time realistic real-wo rld applications, A key component includes voice-signal separation and reco very from a mixture in practical environments. The signal separation and ex traction component includes several algorithmic modules with a variety of s ophistication levels, which include dynamic processing neural networks in t andem with (dynamic) adaptive methods. These adaptive methods make use of o ptimization theory subject to the dynamic network constraints to enable pra ctical algorithms, The underlying technology platforms used in the compiled VOX software can significantly facilitate the embedding of speech recognit ion into many environments. Two demonstrations are described: one is PC-bas ed and is near-realtime, the second is digital signal processing based and is real time. Sample results are described to quantify the performance of t he overall systems.