Industrial Technology Research Institute

Tai-Huei Huang

Shun-Ju Chen

The present invention provides an energy feature extraction method for noisy speech recognition. At first, noisy speech energy of an input noisy speech is computed. Next, the noise energy in the input noisy speech is estimated. Then, the estimated noise energy is subtracted from the noisy speech energy to obtain estimated clean speech energy. Finally, delta operations are performed on the log of the estimated clean speech energy to determine the energy derivative features for the noisy speech.

Energy feature extraction method for noisy speech recognition

The present invention provides a method and a system for determining reliable speech recognition coefficients in noisy environment, which can increase the recognition rate in a noisy environment, such as an in-car environment. The present invention utilizes the feature that most of the energy of in-car noise is concentrated in the low frequency band. Therefore, the input speech signal is filtered to remove the signal at the frequency range in which the noise energy is concentrated. Then, the energy contour of the speech signal is calculated, so as to determine the related speech recognition coefficients. Accordingly, the influence caused by the noise can be reduced, and the recognition rate for the noisy speech can be improved.

Growing community of inventors

Yunlin Hsien, Taiwan

Tai-Huei Huang