Proc. of the Forum Acusticum Sevilla, 2002
Temporal envelope modulation using syllable search method for robust language identification
T. Aoki, M. Komatsu, T. Arai and Y. Murahara
Abstract: Humans are quite capable of identifying languages under noisy conditions while computers still struggle. Robust automatic language identification systems are needed, because there is no place that is totally silent. In this study, multi-layer perceptron is applied using Temporal Envelope Modulation (TEM), a speech signal with reduced spectral information, which is similar to a speech signal in a noisy environment. The experiment used a new method for feature extraction, the Syllable Search Method (SSM), and as a result, the identification rate increased as the number of bands of TEM increased from one to four.