| 摘要 |
A voice recognition system for time encoded speech (TES). The TES coded stream is divided into time frames. Each time frame is fed to a number of seeker circuits, each seeker circuit being optimised to detect an acoustic event of the voice signals, such as voiced or unvoiced events, amplitude or silence (ambiant noise). Segmentation boundaries in the TES symbol stream are then determined from the acoustic events sensed as present or absent. These events are compared with stored events to determine the voice signals. By using this parallel comparison technique comparison is made using distinct boundaries determined from the voice signal content and not by arbitrary boundaries which may contain acoustic events from, for example, more than one word in the voice signals. The implementation of a TES recogniser in low cost CMOS technology makes this parallel comparison technique feasible at reasonable cost. |
