Speech Feature Analysis Using Temporal Linear Embedding

This paper describes a novel speech feature analysis method that creates a readable pattern based on temporal linear embedding. LLE is an unsupervised learning algorithm for feature extraction. If the speech variability is described by a small number of continuous features, then we can imagine the data as lying on a low dimensional manifold in the high dimensional space of speech waveforms. The goal of feature extraction is to reduce the dimensionality of the speech signal while preserving the informative signatures. According to LLE, Temporal linear embedding introduces a time enhanced mechanism to improve system performance. The method remedies the defect that LLE ignores temporal factor which is extremely important for speech signal. In this paper we have present results from the analysis of speech data using PCA and temporal linear embedding. And we observed that the nonlinear embeddings of temporal linear embedding separated certain Chinese phonemes better than the linear projections of PCA.