Speech recognition method, speech recognition device and electronic equipment based on sparse neural network
A neural network and speech recognition technology, applied in the information field, can solve the problems of large scale of neural network, difficult embedded devices or mobile devices, and high cost, and achieve the effect of short training time and reduced scale
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0031] Embodiment 1 of the present application provides a speech recognition method based on a sparse neural network, which is used to recognize a speech segment to be recognized, so as to determine a text corresponding to the speech segment to be recognized.
[0032] figure 1 It is a schematic diagram of the speech recognition method of embodiment 1, such as figure 1 As shown, the method includes:
[0033] S101. Process the speech segment to be recognized to obtain a feature vector of each speech frame in the speech segment to be recognized;
[0034] S102. Using a sparse neural network to identify the feature vector to obtain a state label value corresponding to the feature vector, wherein the weight matrix W of the sparse neural network is obtained based on dimension transformation; and
[0035] S103. Use a decoding model to decode the state tag value to obtain text corresponding to the speech segment to be recognized.
[0036] In this embodiment, speech recognition is pe...
Embodiment 2
[0060] In Embodiment 2, the method for training the weight matrix W based on dimension transformation is described, and the weight matrix W obtained according to the method of this embodiment is used for the sparse neural network adopted in step S102 of Embodiment 1. middle.
[0061] image 3 is a schematic diagram of the method for obtaining the weight matrix W through training in Example 2, such as image 3 As shown, the method includes:
[0062] S301. For the first predetermined number of training speech frames, calculate the Hessian matrix (hessian) of the feature vectors of each training speech frame and the first gradient of the feature vectors of each training speech frame in the first space, and, Based on the first current weight matrix Wm of the sparse neural network in the first space, calculate the state label value corresponding to the feature vector of each training speech frame;
[0063] S302. Project the first current weight matrix Wm and the first gradient f...
Embodiment 3
[0101] This embodiment provides a speech recognition device based on a sparse neural network, corresponding to the speech recognition methods in Embodiment 1 and Embodiment 2.
[0102] Figure 5 is a schematic diagram of the speech recognition device of this embodiment, such as Figure 5 As shown, the speech recognition device 500 includes: a first processing unit 501 , a first recognition unit 502 and a first decoding unit 503 .
[0103] Wherein, the first processing unit 501 is used to process the speech segment to be recognized, so as to obtain the feature vector of each speech frame in the speech segment to be recognized; the first identification unit 502 uses a sparse neural network to identify the feature vector , to obtain a state label value (state id) corresponding to the feature vector, wherein the weight matrix of the sparse neural network is obtained based on dimension transformation; the first decoding unit 503 uses a decoding model to decode the state label valu...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More