Speech recognition method, apparatus, and computer readable storage medium

A speech recognition device, comprising:

• a memory, for storing a program to achieve a speech recognition;
• and
• a processor, for executing the program stored in the memory to achieve the speech recognition by performing the operations of:
• performing a framing and an acoustic feature extraction of a speech-information-to-be-tested according to a default rule to obtain a frame-level speech feature sequence;
• dividing the frame-level speech feature sequence into n blocks sequentially, wherein n is an integer greater than or equal to 2, and each block includes a master block, and each master block is coupled to a slave block, and both of the master block and the corresponding slave block include a plurality of frame speech feature information, and a last-frame speech feature information of the master block is disposed adjacent to a first-frame speech feature information of the slave block;
• inputting all blocks into a preset bidirectional LSTM-RNN model parallelly to compute a forward recognition result of a corresponding neuron in a hidden layer of the preset bidirectional LSTM-RNN model of the master block in each block, and a backward recognition result of a corresponding neuron in a hidden layer of the preset bidirectional LSTM-RNN model of each block;
• and
• obtaining an output result of the corresponding neuron in an output layer of the preset bidirectional LSTM-RNN model corresponding to the forward recognition result and backward recognition result of each block to obtain a speech recognition result of the speech-information-to-be-tested;
• wherein, the preset bidirectional LSTM-RNN model comprises an input layer composed of the corresponding neurons of all blocks, a hidden layer composed of a nonlinear mapping unit, and an output layer composed of the neuron corresponding to the output result of the hidden layer.

A non-transitory computer readable storage medium, for storing one or more programs executed by one or more processors to implement the steps of:

• performing a framing and an acoustic feature extraction of a speech-information-to-be-tested according to a default rule to obtain a frame-level speech feature sequence;
• dividing the frame-level speech feature sequence into n blocks sequentially, wherein n is an integer greater than or equal to 2, and each block includes a master block, and each master block is coupled to a slave block, and both of the master block and the corresponding slave block include a plurality of frame speech feature information, and a last-frame speech feature information of the master block is disposed adjacent to a first-frame speech feature information of the slave block;
• inputting all blocks into a preset bidirectional LSTM-RNN model parallelly to compute a forward recognition result of a corresponding neuron in a hidden layer of the preset bidirectional LSTM-RNN model of the master block in each block, and a backward recognition result of a corresponding neuron in a hidden layer of the preset bidirectional LSTM-RNN model of each block;
• obtaining an output result of the corresponding neuron in an output layer of the preset bidirectional LSTM-RNN model corresponding to the forward recognition result and backward recognition result of each block to obtain a speech recognition result of the speech-information-to-be-tested;
• wherein, the preset bidirectional LSTM-RNN model comprises an input layer composed of the corresponding neurons of all blocks, a hidden layer composed of a nonlinear mapping unit, and an output layer composed of the neuron corresponding to the output result of the hidden layer.