On the use of nearest feature line for speaker identification

Abstract

As a new pattern classification method, nearest feature line (NFL) provides an effective way to tackle the sort of pattern recognition problems where only limited data are available for training. In this paper, we explore the use of NFL for speaker identification in terms of limited data and examine how the NFL performs in such a vexing problem of various mismatches between training and test. In order to speed up NFL in decision-making, we propose an alternative method for similarity measure. We have applied the improved NFL to speaker identification of different operating modes. Its text-dependent performance is better than the dynamic time warping (DTW) on the Ti46 corpus, while its computational load is much lower than that of DTW. Moreover, we propose an utterance partitioning strategy used in the NFL for better performance. For the text-independent mode, we employ the NFL to be a new similarity measure in vector quantization (VQ), which causes the VQ to perform better on the KING corpus. Some computational issues on the NFL are also discussed in this paper.

Cluster-Based Nearest-Neighbour Classifier and Its Application on the Lightning Classification

“…NFL shows good performance in many applications, including face recognition [16] , audio retrieval [17] , image classification [18] , speaker identification [19] and object recognition [20] . The authors of NFL explain that a feature line provides information about the possible linear variants of two sample points not covered by themselves.…”

Robust coarse-to-fine face recognition method

“…[11] interpret that a feature line gives information about the possible linear variants of two sample points. NFL gains good performance in many applications including face recognition [12][13][14][15], audio retrieval [16], speaker identification [17], image classification [18], object recognition [19] and pattern classification [20]. The authors of NFL interpret that the feature line can give information about the possible linear variants of the corresponding two samples very well.…”

“…. Use the s-classes-model to compute the novel contribution of each class in expressing the test sample by formula(17). 5.…”

Pattern classification based on k locally constrained line

“…In NFL, it extends any pair of available prototypes in one class to a feature line (FL). Successful applications of this algorithm include face recognition (Li and Lu 1999), image classification , audio retrieval (Li 2000), speaker recognition (Chen et al 2002), iris recognition (Ma et al 2002), prediction of protein locations (Gao and Wang 2005) and object recognition (Chen and Chen 2004a), etc. As a powerful tool, NFL is also embedded in other learning techniques to obtain better classification accuracy, e.g., subspace learning (He et al 2008;Pang et al 2007Pang et al , 2009, dissimilarity-based classification (Orozco-Alzate et al 2009).…”