ROBUST SPEECH PROCESSING FROM NOISY SPEECH MODELS

摘要

A speech processing system, such as a speech recognition or speech coding system, is capable for processing a degraded speech input signal. The system includes an input for receiving the degraded speech input signal. Means (420) are used for estimating a condition, such as the signal-to-noise ratio or bandwidth, of the received input signal. Means (430) are used means for selecting a processing model which corresponds to the estimated signal condition. The model may be retrieved from a storage (440) with models for different signal conditions. Means (430) are also operable to estimate an originally uttered speech signal based on the received input signal and to process the estimated original signal according to the selected model.