Beamforming method and system based on time-frequency masking value estimation

Abstract

The invention belongs to the technical field of speech enhancement, and in particular relates to a beamforming method and system based on time-frequency masking value estimation. Features The multi-channel speech spectrum feature sequence is obtained through logarithmic transformation, which is sent to the pre-trained and optimized neural network model to obtain the complex-valued time-frequency masking value; the complex-valued time-frequency masking value is converted into the probability of speech existence, and the time-frequency masking is obtained by using the probability model value; calculate the covariance matrix of the speech signal from the time-frequency masking value and the multi-channel speech eigensequence, and perform eigenvalue decomposition on the covariance matrix to obtain the beamforming filter coefficients; combine the beamforming filter coefficients, use the beamforming filter to The speech sequence is processed by speech feature filtering to obtain an enhanced speech signal. The invention integrates the neural network and the spatial clustering to estimate the time-frequency masking value, and improves the performance of beam forming and voice recognition.

Description

technical field [0001] The invention belongs to the technical field of speech enhancement, and in particular relates to a beamforming method and system based on time-frequency masking value estimation. Background technique [0002] Speech coding and speech recognition research is often carried out under laboratory conditions, that is, in environments with high signal-to-noise ratios or in noise-free environments. Therefore, when speech processing moves from the laboratory to practical application, many methods cannot be used due to the existence of actual environmental noise and interference, and the performance drops rapidly. Therefore, it is a practical problem that must be solved to study the processing of improving the auditory effect or improving the signal-to-noise ratio of the noise-reduced speech. The essence of speech enhancement is speech noise reduction. In other words, in daily life, the speech collected by the microphone is usually "polluted" speech with differ...

AI Technical Summary

Problems solved by technology

In the existing speech enhancement processing, the time-frequency masking value estimation based on the neural network has the problem of training-test data mismatch, which affects the performance, and the time-frequency masking value estimation based on the spatial domain clustering has the problem of long iteration time. The value masking value uses the amplitude information of the feature, but does not make full use of the phase information in the feature. The estimation accuracy of the probability of speech and noise needs to be improved.

Images