Far-field speech recognition enhancement method for intelligent water dispenser
A technology of speech recognition and water dispenser, applied in speech analysis, instruments, etc., can solve problems such as poor consistent effect, achieve the effect of improving effect, ensuring reliability, efficient noise and reverberation
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0061] An embodiment of the present invention provides a far-field speech recognition enhancement method for an intelligent water dispenser, figure 1 It is a flow chart of a far-field speech recognition enhancement method for smart water dispensers in an embodiment of the present invention, please refer to figure 1 , the method includes the following steps:
[0062] Step S101, using a microphone array to acquire multi-channel far-field voice signals;
[0063] Step S102, using a Wiener filter to perform noise reduction preprocessing on the speech signal;
[0064] Step S103, obtaining the variance of the expected speech spectrum and the noise-free reverberation spectrum and the expected speech signal spectrum based on the deep learning of the long-short-term memory network;
[0065] Step S104, determine the coefficient of predictive filter to the variance of the expected speech spectrum of output based on WPE algorithm;
[0066] Step S105, determining the desired speech signa...
Embodiment 2
[0075] On the basis of Embodiment 1, the deep learning based on the long short-term memory network obtains the variance of the desired speech spectrum, including:
[0076] The speech signal after the noise reduction preprocessing is subjected to frame division processing;
[0077] Using the log magnitude spectrum of the current frame and adjacent frames as the input signal of the long short-term memory network;
[0078] Outputting the corresponding ideal masking value through the long short-term memory network;
[0079] determining a noise-free reverberation spectrum and an expected speech signal spectrum according to the ideal masking value;
[0080] A variance of the desired speech spectrum is determined from the desired speech signal spectrum.
[0081] The working principle of the above-mentioned technical solution is: the solution adopted in this embodiment is the process of obtaining the variance of the desired speech spectrum through deep learning based on the long-sho...
Embodiment 3
[0084] On the basis of Embodiment 2, the determination of the noise-free reverberation spectrum and the desired speech signal spectrum according to the ideal masking value includes:
[0085] multiplying the noise-reduction preprocessed speech signal by the ideal mask corresponding to the noise-free reverberation spectrum to obtain the noise-free reverberation spectrum;
[0086] The desired speech signal spectrum is obtained by multiplying the noise-reduced preprocessed speech signal by an ideal mask corresponding to the desired speech signal spectrum.
[0087] The working principle of the above-mentioned technical solution is: the solution adopted in this embodiment is the process of determining the noise-free reverberation spectrum and the expected speech signal spectrum according to the ideal masking value, specifically, the speech signal after the noise reduction preprocessing is multiplied by The ideal mask corresponding to the noise-free reverberation spectrum is obtained...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More