Training method and system for child speech recognition model

A speech recognition model and speech training technology, applied in speech recognition, speech analysis, TV system components, etc., can solve problems such as insufficient data volume of children's voices, poor effect of children's speech recognition models, etc., and achieve high recognition accuracy Effect
CN110706692AActive Publication Date: 2020-01-17AISPEECH CO LTD
7 Cites 12 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
AISPEECH CO LTD
Publication Date
2020-01-17

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The embodiment of the invention provides a training method of a child speech recognition model. The method comprises the steps of obtaining training data; training through a baseline acoustic model toobtain an unconditional generative adversarial network; inputting the random noise data into an unconditional generative adversarial network to obtain noise enhanced acoustic features; inputting thenoise enhancement acoustic features into a baseline acoustic model to obtain a posterior probability soft label corresponding to each frame of noise enhancement acoustic features; and training the children speech enhancement acoustic recognition model at least by taking the noise enhancement acoustic features, the soft label, the children speech training data and the hard label as sample trainingdata. The embodiment of the invention further provides a training system of the child speech recognition model. According to the embodiment of the invention, the pronunciation nature of the child speech is changed under the condition that the child speech is limited, diversified child speeches are generated, and the recognition accuracy of the child speech recognition model is improved.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The invention relates to the field of speech recognition, in particular to a training method and system for a children's speech recognition model. Background technique

[0002] With the development of intelligent voice, it not only provides a large number of voice interactive products for adult users, but also provides many intelligent products for children, such as intelligent story machines and intelligent robots. However, due to the differences between children's voices and adult voices, existing speech recognition systems are not effective in recognizing children's voices.

[0003] For the above problems, the method of adding noise is usually used: in data preprocessing, for each sentence in the clean audio of children's voices, use the FaNT tool to add random one of 115 noises with a signal-to-noise ratio of 20dB to increase the number of children's voices amount of data; or, adopt random feature mapping method: random feature mapping learns an a...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More