基于LVQ的普米语语谱图识别

Primi Language Spectrogram Recognition Based on Learning Vector Quantization

摘要: 本文使用基于学习向量量化算法实现了对普米语语谱图的识别。该算法首先通过傅立叶变换绘制出每条普米语语料的语谱图；再次，提取每张语谱图的图像灰度共生矩构建特征向量；最后，运用学习向量量化实现对普米语孤立词的分类。基于学习向量量化算法的分类准确率达到了80.0%。

Abstract: In this paper, we proposed a Primi language speech recognition algorithm based on Learning Vector Quantization (LVQ). Firstly, the algorithm uses the Fourier transform to plot each Primi language pragmatics; secondly, it extracts the gray level cooccurrence moments of each spectrogram to construct feature vectors; and finally, it uses learning vector quantization to realize the classification of Primi isolated words. The classification accuracy based on the Learning Vector Quantization (LVQ) is 80.0%.