Acoustic model training method and device

Abstract

The embodiment of the invention provides an acoustic model training method and device. The method includes the steps of establishing a deep neural network model initial model; dividing voice training data into N non-intersecting data subsets, renewing the deep neural network model initial model for each data subset by means of a stochastic gradient descend algorithm to obtain N deep neural network model sub models, wherein N is an integer larger than or equal to 2; integrating the N deep neural network model sub models to obtain a deep neural network model intermediate model, and judging that the deep neural network model intermediate model is a trained acoustic model when the deep neural network model intermediate model conforms to preset convergence conditions. By means of the acoustic model training method and device, training efficiency of the acoustic model is improved, and performance of voice recognition is not reduced.

Description

technical field [0001] The embodiments of the present invention relate to the technical field of speech recognition, and more specifically, to an acoustic model training method and device. Background technique [0002] Speech recognition is a technology that converts speech signals into text. It is a convenient way of human-computer interaction and is now widely used in mobile Internet and other fields. Speech recognition is a serialized classification problem, the purpose of which is to convert a series of collected speech signals into a series of text outputs. The fields involved in speech recognition technology include: signal processing, pattern recognition, probability theory and information theory, sound mechanism and auditory mechanism, artificial intelligence and so on. [0003] Traditional speech recognition systems are generally divided into three modules, namely: acoustic model, such as the model described by the HMM-GMM system framework; language model, such as ...

AI Technical Summary

Problems solved by technology

However, in big data, the second-order parameter optimization method often requires a lot of detailed adjustments. In the absence of prior knowledge, the second-order optimization is often not as robust as the first-order optimization.

Specifically, for DNN modeling in speech recognition, this algorithm does not achieve the good performance of the SGD algorithm

Images