Voice Preprocessor for Digital Voice Applications.

摘要

A voice processor operating satisfactorily in laboratory environments with carefully prerecorded speech samples often fails to operate satisfactorily with live speech. Potential reasons are: the speech level may be too high or too low; the speech signal may have too much interference (ambient noise, breath noise, 60 Hz hum, digital noise in analog circuits, a DC bias (caused by component aging, etc.) generated at the analog-to-digital converter output); the microphone frequency may be severely distorted; the speech signal from the existing audio system, in certain operating environments, may be improperly coupled to the front-end circuit; the speaker may be talking too fast or may have an improper mouth-to-microphone distance, or the speech may lack high-frequency energies. In this report, we have generated a comprehensive design for a speech preprocessor that removes interferences, adaptively equalizes frequency anomalies, and conditions speech for speech encoding, speech recognition, speaker recognition, or extraction of verbal or nonverbal information from speech. Speech preprocessing, Speech conditioning, Perception of speech distortion, Microphone response equalization, Automatic gain control, Digital antialiasing filtering. (jes)