Method, apparatus and storage medium for constructing speech decoding network in numeric speech recognition

摘要

A method for constructing a speech decoding network in numeric speech recognition. The method is an operation of acquiring training data acquired using a numerical audio recording, the training data including a plurality of audio segments, and each audio segment includes an operation (110) including a plurality of audio numbers. In order to obtain a feature sequence corresponding to each speech segment, an acoustic model according to an operation (130) for performing acoustic feature extraction on training data, a feature sequence, and a phoneme corresponding to a number in the training data In order to obtain a speech decoding network, an operation (150) of performing a progressive training starting from a monophonic acoustic model, a language model is obtained, and the language model and the acoustic model obtained through the training are used to establish a speech decoding network. And an operation to construct (170).