Acoustic model training and constructing method, acoustic model and speech recognition system

Abstract

The invention provides an acoustic model training and constructing method, a hidden Markov acoustic model based on the training method, and a speech recognition system. The training method comprises the following steps: (1) calculating the frame statistical number of each class and an intra-class divergence matrix based on training data and a pre-given state cluster; (2) for a non-speech state class in a model, inhibiting and smoothing the statistical number of the state class if the frame statistical number corresponding to the state class is much higher than the average statistical number of state classes; (2) for a speech state class in the model, inhibiting and smoothing the statistical number of the state class if the frame statistical number corresponding to the state class is much lower than the average statistical number of state classes; (4) calculating a heteroscedastic linear discriminant analysis matrix based on the intra-class divergence matrix and the smoothed class statistical number; and (5) using the calculated heteroscedastic linear discriminant analysis matrix in speech characteristic and model dimension reduction, and carrying out iteration again to get a dimension-reduced stable acoustic model. The recognition performance of the acoustic model is improved eventually.

Description

technical field [0001] The invention belongs to the field of speech recognition, and in particular relates to a smoothing method for heteroskedasticity linear discriminant analysis, which can be used for fast dimensionality reduction and decorrelation processing of high-dimensional feature vectors in language recognition. Background technique [0002] In large-vocabulary continuous speech recognition, heteroscedastic linear discriminant analysis (HLDA, Heteroscedastic Linear Discriminant Analysis) improves the recognition performance of the model by removing the correlation between features, which is widely used in acoustic modeling (N.Kumar. Investigation of silicon auditory models and generalization of linear Discriminant analysis for improved speech recognition.PhDthesis, Johns Hopkins University, Baltimore, Maryland, 1997.). The core of its algorithm is to divide the speech into different classes according to the state, and reduce the dimensionality of the original featu...

AI Technical Summary

Problems solved by technology

However, in practical applications, it is found that, on the one hand, when used for speech recognition, the state class describing silence usually occupies a large amount of statistics, far exceeding the single state class describing speech, which makes the heteroscedasticity linearity calculated based on statistics Discriminant analysis, which is too biased towards silence, inhibits the distinction of speech parts to a certain extent; on the other hand, due to limited training data, the state distribution of some speech is relatively sparse, and the corresponding training data cannot reflect its true statistical distribution. , which leads to the statistical information in the calculation of heteroscedastic linear discriminant analysis is also affected accordingly

Images