| 摘要 |
PURPOSE: A method for estimating a noise mask using a hidden Markov model and an apparatus for performing the same are provided to improve estimation accuracy of noise mask information and cancel noise using noise mask information with improved estimation accuracy, thereby improving voice recognition performance. CONSTITUTION: A method for estimating a noise mask using a hidden Markov model includes an auditory characteristic modeling step, a direction parameter obtaining step, and a noise mask information estimating step. In the auditory characteristic modeling step, an audio signal with noise is converted into an auditory frequency signal and envelope information(301). In the direction parameter obtaining step, a direction parameter configured by a mutual size difference and a mutual time difference is obtained. In the noise mask information estimating step, noise mask information is estimated based on a noise mask model based on a hidden Markov model learned by using the mutual size difference and the mutual time difference as feature vectors.
|
