METHOD AND APPARATUS USING HARMONIC MODEL-BASED FRONT END FOR ROBUST SPEECH RECOGNITION

摘要

provides a system and method for reducing noise in a speech signal . The system and method decompose a noisy speech signal as a harmonic component and a residual component . Then, the harmonic component and residual component are combined to form a noise reduced value . In some embodiments , this sum is the harmonic component is a weighted sum multiplied by a scaling factor . In some embodiments , the noise reduction value is used for speech recognition .