Speech recognition method based on neural network model

Abstract

The invention discloses a speech recognition method based on a neural network model. The speech recognition method comprises the steps of: acquiring a training sample set, extracting first feature parameters of the training sample set, acquiring the neural network model which can be used for recognition, and establishing a template library; and acquiring a prediction sample set, inputting the prediction sample set into the neural network model, extracting to obtain second feature parameters, and performing matching prediction on the second feature parameters and the first feature parameters inthe template library to obtain an identification result, wherein a neural network comprises a convolutional neural network and a recurrent neural network which are arranged in sequence. Compared withthe prior art, the speech recognition method provided by the invention has the advantages that the convolutional neural network and the recurrent neural network are effectively combined and connected, so that the overall learning efficiency and robustness of the network can be improved and the speech recognition performance can be improved while the speech recognition accuracy rate is ensured.

Description

technical field

[0001] The invention relates to the field of speech recognition, in particular to a speech recognition method based on a neural network model. Background technique

[0002] People in today's world have already used and relied on machines to an extremely high level. On this basis, people have also begun to desire more intuitive and easier operations on machines to achieve their goals. If they can directly use sound transmission In the form of direct control of the machine to help you accomplish what you want to do, many things will get twice the result with half the effort, so the operation of speech recognition technology was born. The essence of speech recognition technology is to process and classify the information contained in the sound. People can use speech recognition to make the machine obtain the information contained in the sound. The birth of this technology simplifies some operating commands to the machine, liberates human hands, and greatly faci...

Examples

Embodiment 1

[0038] Such as figure 1 As shown, this embodiment includes: obtaining a collection of training samples, preprocessing the collection of training samples, and converting the collection of training samples into an MFCC characteristic parameter matrix; extracting the first characteristic parameter of the collection of training samples, and inputting the first characteristic parameter into the neural network The network is trained, and the error between the output value of the neural network and the original signal label is compared; the weight and bias inside the neural network are continuously updated to obtain a neural network model that can be used for recognition, and a template library is established, which includes different speech The first feature parameter of the prediction sample set; the prediction sample collection is obtained, the prediction sample collection is preprocessed, the prediction sample collection is converted into an MFCC feature parameter matrix, and inpu...

technical field

[0001] The invention relates to the field of speech recognition, in particular to a speech recognition method based on a neural network model. Background technique

[0002] People in today's world have already used and relied on machines to an extremely high level. On this basis, people have also begun to desire more intuitive and easier operations on machines to achieve their goals. If they can directly use sound transmission In the form of direct control of the machine to help you accomplish what you want to do, many things will get twice the result with half the effort, so the operation of speech recognition technology was born. The essence of speech recognition technology is to process and classify the information contained in the sound. People can use speech recognition to make the machine obtain the information contained in the sound. The birth of this technology simplifies some operating commands to the machine, liberates human hands, and greatly faci...

Examples

Embodiment 1

[0038] Such as figure 1 As shown, this embodiment includes: obtaining a collection of training samples, preprocessing the collection of training samples, and converting the collection of training samples into an MFCC characteristic parameter matrix; extracting the first characteristic parameter of the collection of training samples, and inputting the first characteristic parameter into the neural network The network is trained, and the error between the output value of the neural network and the original signal label is compared; the weight and bias inside the neural network are continuously updated to obtain a neural network model that can be used for recognition, and a template library is established, which includes different speech The first feature parameter of the prediction sample set; the prediction sample collection is obtained, the prediction sample collection is preprocessed, the prediction sample collection is converted into an MFCC feature parameter matrix, and inpu...

Images